Hindawi

Scientific Programming
Volume 2017, Article ID 8093057, 12 pages
https://doi.org/10.1155/2017/8093057

Research Article
Fashion Evaluation Method for Clothing Recommendation
Based on Weak Appearance Feature

Yan Zhang,1,2 Xiang Liu,1 Yunyu Shi,1 Yunqi Guo,3 Chaoqun Xu,3 Erwen Zhang,4
Jiaxun Tang,1 and Zhijun Fang1
1
School of Electronic and Electrical Engineering, Shanghai University of Engineering Science, Shanghai 201620, China
2
School of Fashion Technology, Shanghai University of Engineering Science, Shanghai 201620, China
3
School of Urban Rail Transportation, Shanghai University of Engineering Science, Shanghai 201620, China
4
School of Mechanical Engineering, Shanghai University of Engineering Science, Shanghai 201620, China

Correspondence should be addressed to Xiang Liu; [email protected]

Received 3 March 2017; Accepted 2 May 2017; Published 1 October 2017

Academic Editor: Chi-Hung Chi

Copyright © 2017 Yan Zhang et al. This is an open access article distributed under the Creative Commons Attribution License,
which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

With the rapid rising of living standard, people gradually developed higher shopping enthusiasm and increasing demand for
garment. Nowadays, an increasing number of people pursue fashion. However, facing too many types of garment, consumers need
to try them on repeatedly, which is somewhat time- and energy-consuming. Besides, it is difficult for merchants to master the
real-time demand of consumers. Herein, there is not enough cohesiveness between consumer information and merchants. Thus, a
novel fashion evaluation method on the basis of the appearance weak feature is proposed in this paper. First of all, image database
is established and three aspects of appearance weak feature are put forward to characterize the fashion level. Furthermore, the
appearance weak features are extracted according to the characters’ facial feature localization method. Last but not least, consumers’
fashion level can be classified through support vector product, and the classification is verified with the hierarchical analysis method.
The experimental results show that consumers’ fashion level can be accurately described based on the indexes of appearance weak
feature and the approach has higher application value for the clothing recommendation system.

1. Introduction depends on manual operation. To be specific, salesmen need

to recommend garment to customers in order to arouse
The fashion industry occupies a significant position in the their interest in purchasing. However, it is very difficult for
global economy and involves large industrial chain, including salesmen to understand customers’ real thoughts and then
garment design, production, and sales. In fact, in the recent recommend the targeted garment as there is no sufficient
years, there has been an expanding demand for clothing all cohesiveness between customer information and merchants.
over the world. Since 2008, the garment sales have increased Therefore, it is essential and meaningful to find a set of objec-
by $3.3 billion every year, and the global garment sales tive indicators, instead of subjective opinions, to evaluate the
reached $1.25 trillion in 2012 [1]. According to a report of fashion level in the clothing recommendation technology.
Euromonitor International, in 2015, the growth rate of cloth- As the Internet technology continues to develop rapidly,
ing sales was 4.5%, and the industry gross reached $1.6 virtual fitting and other clothing intelligent equipment have
trillion. The global clothing sales enhanced by 3.8% and the enjoyed great popularity in the fashion industry. Cordier et al.
industry gross rose to $1.7 trillion in 2016. The above data (2003) [2] first applied the 3D graphics technology to create
show that the garment industry is developing at a rapid rate. and simulate the virtual store. Subsequently, Li et al. (2011)
In fashion sales, the recommendation technology, as an [3] proposed the interactive 3D virtual fitting room system,
emerging technology, has attracted wide attention of scholars. in which the model’s hairstyle and accessories can be changed
As is widely known, the traditional garment recommendation according to customers’ preferences and customers’ matching
2 Scientific Programming

Eyebrows
Data Make-up Eye shadow
collection Cheek color
Lip color Support vector Test
result Fashion level
Neck/hand accessories machine
Database Extract
appearance Brooch
establishment Accessories
weak feature Nail Verification
Hat Comparison with
Red Green Analytic garment level
Input Test
image Hair color Blue Brown hierarchy
Gray process result
White
Black Yellow
Garment
Ground truth Ground truth Image recommendation
Compare
eatablishment scoring scoring result

Figure 1: Process of our approach.

degree can be evaluated to guide people to choose the suitable system based on the principles of wearing fashion and
clothes. Nevertheless, the virtual fitting research products are aesthetic is studied [9]. In addition to the above work, Iwata et
constantly innovating and developing. In fact, today’s systems al. (2011) [10] offered a recommender system, utilizing fashion
are mainly used to display garment, and customers can only magazines’ full-body photographs. In the same way, Sha et al.
have a preview of the fitting effect. If the store does not have (2016) [11] extracted multiple features from images to analyze
an efficient recommended method, the search will be tedious their contents in different attributes, such as fabric pattern,
and frustrating. Zhang et al. (2008) [4] presented an inter- collar, and sleeve. Some garment system integrates the fashion
action clothes fitting system that can recognize what human themes and shapes professional designers’ knowledge and
eyes perceive in terms of the clothing similarity through perception to help them choose the most relevant garment
the frontal-view outfit images. Limaksornkul et al. (2014) design scheme for a specific customer [12].
[5] put forward the Closet Application to record the clothing In the computer realm, the concept of the recommenda-
statistics and accessories that are frequently used to recom- tion technology was first introduced in the middle 1990s [13].
mend clothes to customers according to the statistics of their So far, different advanced algorithms have been developed.
purchasing history. The recommended technology not only The following is a review of the relevant methods. The first
allows customers to quickly find the right clothes in the fitting method is the content-based recommendation algorithm.
process but also helps businesses increase sales. Nonetheless, For example, the CRESA combined textual attributes, visual
the above methods are mainly based on the subjective views features, and human visual attention to compose the clothes
that ignore the objective data. To address this problem, this profile in the recommendation [14]. Ajmani et al. (2013) [15]
paper proposes a fashion level evaluation method for clothing present a novel method for content-based recommendation
recommendation based on the weak appearance feature. of media-rich commodities with the use of probabilistic
This paper is organized as follows. Section 2 analyzes multimedia ontology. Li et al. (2012) [16] utilized the HMM of
and summarizes the current situation of recommendation recommended items to match customers’ model according to
technologies as well as their advantages and disadvantages. customer data. The second method is the collaborative
Section 3 (Figure 1) contains the description of the definition filtering-based recommendations algorithm. For instance,
of fashion, classifies the fashion index, extracts the weak Nogueira et al. (2015) [17] presented a new collaborative fil-
features of human appearance, and describes the SVM clas- tering strategy that utilizes the visual attention to characterize
sification. The experimental method and the result of the images and alleviate the new item cold-start problem. The
experimental analysis are given in Section 4. Last but not rule-based recommendation algorithm is the third method.
least, Section 5 presents the conclusion and summarizes the Hwang et al. (2016) [18] put forward a method to generate the
contents of this paper. automatic rules with the user’s items and made a suggestion
on the best rule. The fourth method is the utility-based
recommendation. For instance, Scholz et al. (2015) [19] found
2. Related Works that exponential utility functions are better geared to pre-
There are lots of methods achieved in garment recommenda- dicting optimal recommendation ranks for products, and
tion. For instance, customer ratings and clothing are utilized linear utility functions perform much better in estimating
as considerations for garment recommendation [6]. Similarly, customers’ willingness.
user’s personal preference and the history of clothing items
have been tried [7]. Furthermore, some scholars found that 2.1. Conclusions on the Literatures. From the review of the
the past statistics of clothes and accessories and current related literatures, the following conclusions can be drawn.
weather conditions as well as special occasions can provide (1) The recommendation method based on customer
a relevant recommendation on garment [8]. In order to rating and personal interests, to some extent, has backward
meet different needs, an intelligent clothing recommendation features. In practical conditions, most customers would judge
Scientific Programming 3

Table 1: Customer fashion level classification. Table 2: Garment fashion level classification.
Fashion level Description classification Fashion level Description classification
First level Wonderful First level Fashion trend
Second level Great Second level Popular trend
Third level Good Third level Traditional trend
Fourth level Common Fourth level Common trend

the vogue of clothing according to the subjective feelings and Table 3: Weak appearance features catalogue.
matching degree. In particular, most people will become con-
fused when selecting clothes. Actually, it is the representation Category Weak feature index
of customers’ ambiguity regarding their personal conditions. Make-up Eyebrow, blush, lips, eye shadow
(2) As for the recommendation method based on the con- Accessories Neck accessories, hand accessories, brooch, nail, hat
tents, it is applicable for multiple regions. Typically, it will Hair color Red, yellow, green, blue, brown, black, gray, white
recommend new projects to users according to the individual
browsing records. The recommendation results have been
proven to be explicit and accessible. However, the content-
based recommendation method is relatively improper when consider the quarter sales, clothing style, and other factors.
applied in the fashion industry, which can be ascribed to the As a result, garments can be divided into four categories.
data cold-start problem. That is to say, new users without The classification of fashion level is a subjective method
any browsing record could not obtain recommendations. In that needs subjective evaluation on the image characters
addition, it is rather difficult to process clothing products with through the expert group. The subjective method is closely
the relatively complicated attributes. related to the subjective factors, such as knowledge back-
(3) The retrieval means of customer data are of impor- ground and psychological motivation of the experts involved
tance. In the past, businessmen would retrieve customers’ in the evaluation. Actually, this is time-consuming, taking
data through membership cards and questionnaires. Such up lots of resources and causing unstable data classification
method fails to guarantee the authenticity of data, since the results. As for the researches of visual psychological charac-
data could not be timely upgraded in case of any physio- teristics, there has been no quantitative description method
logical and psychological changes of customers. At present, by which the objective evaluation results can represent the
e-commerce businessmen mainly make recommendations subjective evaluation results.
through registering virtual members, tracking consumption, This dissertation aims to find out a set of objective indexes
and browsing information. However, the recommendation which can be used to assess fashion level. The fashion level
results are sometimes not ideal, since some customers doubt is defined as the individual degree of fashion. Fashion level
if their private information will be exposed. Confronted by shows people’s appearance, dress, act, and so forth. Consid-
the difficulty in data retrieval, this paper utilizes the camera ering all the factors that affect the evaluation of personnel’s
as the output equipment of image and video to retrieve cus- scoring, this paper regards the weak appearance feature as
tomers’ data. After building the image database and analyzing an important index that can influence the fashion level. The
the image data, this paper subsequently classifies customers weak appearance feature here means that the individual
according to their fashion level. This classification will be features have low characterization degrees. There are many
upgraded with the change of customers’ data. In the end, weak appearance features related to the individual fashion
garment recommendation will become easy according to level.
customers’ classification data and clothing classification data. Under these conditions, if the camera can recognize the
person (Figure 2), effective image data can be accessed to.
3. Our Proposed Method Three major categories, namely, make-up, accessories, and
hair colors, of the weak appearance feature are extracted,
The word “fashion” is originated from the translation of including eye shadow, blush, lip color, eyebrow color, hat,
VOGUE, a famous US fashion magazine. Fashion level is a accessories on hand and neck, nails, brooches, and hair color.
way of life and the awareness of pursuing the real, good, and By utilizing the SVM classification method, as shown in
beautiful things. Different individuals tend to have different Table 3, fashion level is evaluated based on whether human
pursuits of fashion. This paper characterizes individuals body has weak appearance features. The garment recom-
based on different fashion levels. To quantize the fashion mendation is established on fashion level classification; the
level, customers’ fashion levels are divided into four degrees comparison result of fashion level category and ground truth
(Table 1), namely, wonderful, great, good, and common. Such validate the correctness of fashion level classification results.
classification aims to provide objective clothing recommen- There is no effective way to establish fashion level database.
dations to customers. Nonetheless, the fashion level image database established in
Furthermore, the garment fashion level is classified the paper is the basis of follow-up studies. Effective fashion
(Table 2) based on the data from fashion designers, buyers, image database is of vital significance to the training and
vendors, and producers. The classification process needs to testing of algorithm.
4 Scientific Programming

Analytic
hierarchy process Verification
Comparison

Extract
Ground truth
Input image appearance weak Support vector machine Test result
feature scoring

Garment Comparison with

Fashion level
recommendation garment level

Figure 2: Fashion level classification framework based on weak appearance feature.

3.1. Establishment of Database Table 4: Customer age group classification.

3.1.1. The Establishment of Image Library Age Age group

15–25 First
Image Source. Images are selected from the pictures collected 25–35 Second
through work, study, entertainment, and rest or from the 35–45 Third
Internet. 45–55 Forth

Object Definition (Table 4). The research objects are limited to

Asian women, aged 15∼55, excluding the ones younger than expert group for the same image should be removed, and the
15 and older than 55. average of the rest of the scores is regarded as the standard
scores of the image fashion level.
Photo Regions. The image should cover all the weak feature
regions, including the three types of indicators which are 3.2. The Extraction of Weak Feature Index. The current
make-up, accessories, and hair colors. The images (shown in face detection methods mainly include two categories: the
Figure 3) should contain the frontal face images of the objects, knowledge-based ones and statistics-based ones. To extract
while the images of sides and rear are invalid. The JPG format the weak facial feature, the facial feature points are located
is adopted in image data. first and then the face recognition is started (Figure 4). This
paper adopts the Adaptive Boosting method for facial feature
Image Quality Requirements. The efficiency is greater than positioning. Adaptive Boosting method, proposed by Freund
100 dpi and the requirement of image sharpness is related to (1995) [20], is a statistical learning method that integrates the
the resolution size. The images taken by camera during the weak classifier with the strong classifier [21, 22]. The basic
daytime generally can meet the requirements; in the evening, idea is to endow large weight to the unsuccessful training
shadow occlusion area will be generated under the light. For samples, make learning algorithm focus on the difficult
example, the shadow of the neck area is blocked by the head training samples in the subsequent study, and finally weight
so the images taken in the evening are excluded in the data and add a number of weak classifiers selected by the algo-
source. Image database consists of high-resolution and high- rithm to strong classifier.
quality color source images. The input of the algorithm is a set of data matrix,
(𝑥1 , 𝑦1 ), . . . , (𝑥𝑚 , 𝑦𝑚 ), where 𝑥𝑖 belongs to a sample space 𝑋
Image Grouping. The images are randomly divided into the and 𝑦𝑖 belongs to a sample space 𝑌. This paper assumes that
training group (800 images) and test group (200 images), and 𝑌 = (−1, +1). Subsequently, the learning-based algorithm is
many tests have been conducted for verification. used for 𝑡 times, 𝑡 = (1, . . . , 𝑇), so as to maintain the weight
distribution of training data set. The data are set in the 𝑡 round
3.1.2. The Establishment of Ground Truth weight value of 𝐷𝑡 (𝑖). All the weights at the beginning of the
training are given values. However, at the end of each round,
The Expert Group. It consisted of 60 experts in the fashion the weight value of the data which are wrongly classified
field (20 fashion designers, 20 garment buyers, and 20 gar- will rise. As a result, the weak learning device focuses on
ment salesmen) and 40 experts in other fields (20 nonartistic the classification of a more difficult part. The weak learning
experts and 20 students of nonart major). device is utilized to find the weak assumption ℎ𝑡 , and the
superiority of the weak assumptions is decided by its level of
Expert Scoring. The expert group judges the fashion level from error 𝑒𝑡 .
the perspective of professional knowledge according to the
images under the premise that they are not informed of the 𝑒𝑡 = 𝑃𝑟𝑖∼𝐷𝑖 [ℎ𝑡 (𝑥𝑖 ) ≠ 𝑦𝑖 ] = ∑ 𝐷𝑡 (𝑖) . (1)
feature indexes. The highest and lowest scores offered by the 𝑖:ℎ𝑡 (𝑥𝑖 )=𝑦
̸ 𝑖
Scientific Programming 5

Figure 3: Database images sample.

Figure 4: Locate facial features.

The indexes are judged according to the different char- eyebrows area are assumed as follows: 𝜑1 (𝑥1 , 𝑦1 ), 𝜑2 (𝑥2 , 𝑦2 ),
acteristics of the weak feature indexes and through different 𝜑3 (𝑥3 , 𝑦3 ), and 𝜑4 (𝑥4 , 𝑦4 ). The eyebrows area is set by the
methods or steps. Figure 5 shows weak appearance feature rectangle which is connected by the four boundary points,
identification framework. and the rectangle image is clipped. Secondly, the gray-scale
image processing is conducted to obtain gray image. Later,
the binarization processing is executed on the images. It is
3.2.1. The Make-Up Index Includes Eyebrows, visible that the images of the ones who pencil the eyebrows
Eye Shadow, Cheek Color, and Lip Color show dense and uniform black distribution, while the images
of the ones who do not pencil the eyebrows have sparse and
(1) Eyebrows. It is important to first determine whether uneven black distribution. Since there are obvious differences
consumers pencil eyebrows nor not. The facial images of between the black distribution of the binarization images of
volunteers who pencil the eyebrows and those who do not the ones who pencil the eyebrows and the ones who do not,
pencil the eyebrows are selected, with an image processing the black pixel density, namely, the black area in the rectangle
process shown in Figure 6. First of all, the Laplace operator is area, is calculated. The black pixel density threshold of the
used to detect the eyebrows edge. The four boundary points of eyebrows can be obtained through the statistical analysis, and
6 Scientific Programming

Eyebrows
Make-up Eye shadow
Cheek color
Lip color

Neck/hand accessories
Face Eyeball Brooch
Image Accessories
location location Nail
Hat
Red Green
Hair color Blue Brown
Gray White
Black Yellow

Figure 5: Weak appearance feature identification framework.

Laplace operator
Gray processing Binarization processing
detecting eyebrows edge

(A): image of the 1 (x1 , y1 ) 2 (x2 , y2 ) 1 (x1 , y1 ) 2 (x2 , y2 ) 1 (x1 , y1 ) 2 (x2 , y2 )

pencilled eyebrows 3 (x3 , y3 ) 4 (x4 , y4 ) 3 (x3 , y3 ) 4 (x4 , y4 ) 3 (x3 , y3 ) 4 (x4 , y4 )

(B): image of the not 1 (x1 , y1 ) 2 (x2 , y2 ) 1 (x1 , y1 ) 2 (x2 , y2 ) 1 (x1 , y1 ) 2 (x2 , y2 )
3 (x3 , y3 ) 4 (x4 , y4 ) 3 (x3 , y3 ) 4 (x4 , y4 ) 3 (x3 , y3 ) 4 (x4 , y4 )
pencilled eyebrows

Figure 6: Eyebrows area sample contrast.

Forehead area

Cheek color
detection area

Eye shadow area

Figure 7: Eye shadow.

Figure 8: Cheek color.

it is regarded as the judgment to determine whether people
pencil the eyebrows.

(2) Eye Shadow. As indicated in Figure 7, the rectangle in threshold is determined by many experiments. When the
the middle area of eyes and eyebrows can be defined as the contrast ratio is at range of the threshold, it is considered that
eye shadow area; the area above eyes and the lower edge of one does not use the blusher. Otherwise, it can be considered
hair constitute the forehead area; features are extracted from that one uses the blusher.
the eye shadow and forehead areas. The maximum likelihood
estimates that RGB of the two areas can be compared. If the (4) Lip Color. Figure 9 shows lip structure. In order to
maximum likelihood estimation is almost the same, it can be determine whether consumers apply lipstick, 100 color pho-
concluded that one does not use eye shadow; on the other tographs are selected; lip color samples are chosen; RGB
hand, if it is different, one uses eye shadow. values of four sample points of upper lip bead and lower lip
bead are manually selected. Then, RGB values are converted
(3) Cheek Color. As indicated in Figure 8, the triangle area to HSV value (color, purity, and brightness). The four groups
below eyes is set to be cheek color area; the cheek and fore- of HSV value are calculated. Everyone has four detection
head areas are extracted. The maximum likelihood estimates areas; a total of 12 HSV data value is classified as lip color
that RGB of the two areas can be compared. The contrast features.
Scientific Programming 7

is used with training error as the optimization problem

constraint and with the minimized incredible range as the
optimization goal. SVM is developed from the statistical
Lip color
pattern recognition theory. SVM can be utilized to solve
detection area two kinds of classification problems by finding an optimal
separating hyperplane and separating two classes of data
with maximum intervals mainly based on the structural
risk minimization principle. Finally, solving of a linearly
constrained convex quadratic programming (QP) is involved.
Figure 9: Lip color. Thus, the solution of support vector machine is unique and
optimal [26].
In the SVM training model, 1000 images are selected as
experimental data from database. In the second place, 800
3.2.2. The Accessories Index Includes Neck and groups of data are randomly selected as the training groups
Hand Accessories and Bracelet and Nails and trained for 100 times. The paper selects the appearance
weak features, including eye shadow, cheek color, lip color,
(1) Neck and Hand Accessories. The edge detection technique
eyebrow color, hats, neck and hand accessories, nails, and
can be used to finger the gesture recognition [23]. People’s
brooches and red, yellow, green, blue, brown, black, gray,
wrist, finger edge, and neck edge are detected. If the obvious
and white colors. After determining the SVM model as C-
edge cannot be detected in the areas, it shows that people wear
SVC, the type of kernel function as the radial basis function
neck or hand accessories; otherwise, it can be considered that
(RBF), the relevant punishment coefficient, and the radius
people do not wear neck or hand accessories.
of kernel function, a 12 d linear hyperplane is established to
(2) Brooch. Brooch is highly reflective, and its image area divide samples. Later, the remaining 200 groups of data are
color space has obvious characteristics. Thus, brooch can be used as the validation group and the number of verifications
identified through the extraction of the color space [24]. The is 100 times. Subsequently, the test results are compared with
area below neck and above waist is extracted as the chest the ground truth score and the accuracy of the model is tested.
area. The RGB values of the chest area are extracted, and
then the RGB values are converted to the HSV value. If the 3.4. Validation with the Analytic Hierarchy Process. Analytic
highlighted area is judged through statistical analysis and the hierarchy process (AHP) is used for the validation of SVM
size is between 1 cm2 and 9 cm2 , it can be argued that the algorithm. AHP regards the research object as a system,
people wear brooch. Otherwise, if there is no highlighted area which does not cut out the influence of factors on the results.
or the highlighted area is too small or too large, it is thought The results can be affected by each layer weight setting in
that people do not wear brooch. AHP directly or indirectly. Besides, the influence level of each
factor on the results is clear and quantitative [27]. The specific
(3) Nail. As shown in Figure 10, after machine identifies nails, modeling process is as follows.
the nail color threshold of nails without nail polish can be
obtained through statistical analysis [25]. If consumers’ nail 3.4.1. The Individual Fashion Level of Each Type of Feature Is
color is within the threshold, it can be regarded that they do Calculated Based on AHP. (1) The criterion function of weak
not apply nail polish; if not, it is considered that they apply feature is established for weak feature in the classification
nail polish. of make-up. If customers have the feature, the correspond-
ing function assignment of the feature is “1”; if not, the
(4) Hat. Edge detection is conducted on head position. If Ω corresponding function assignment is “0,” as shown in the
shape or semicircle closed graph is measured, hair color or following formula:
environmental color is extracted for further comparison; if
the color is obviously different, it is argued that consumers {1, with the feature
wear hat; otherwise, it is argued that consumers do not wear ℎ𝑖 = { 𝑖 = 1, 2, 3, 4, (2)
0, without the feature,
hat. {
Through the combined form of expert suggestion and
3.2.3. Hair Color. A polar coordinate system is established extensive survey, matrix 𝐴 1 of the feature in the classification
with two-eyed center as the center of the circle and each line of make-up is given as follows:
extends from the origin. The first detection edge pixel point 1 1
is marked with 𝑀𝑖 ; the second detecting edge pixel point is [1 3
2
2]
marked with 𝑁𝑖 ; the area marked between 𝑀𝑖 and 𝑁𝑖 is hair [ ]
[3 1 4 2]
area. The RGB maximum likelihood estimator of the region [ ]
[ ]
is extracted as the hair color values. 𝐴1 = [ 1 1 1] . (3)
[ 1 ]
[2 4 3]
[ ]
[ 1 ]
3.3. SVM Algorithm. After reading the images in the database 2 3 1
and identifying the weak feature, the support vector product [ 2 ]
8 Scientific Programming

Figure 10: Nail recognition.

The weight of the weak feature is calculated based on the Table 5: RI value change.
feature matrix and the following are the results: 𝜔11 = 0.2289,
𝜔12 = 0.0267, 𝜔13 = 0.6538, and 𝜔14 = 0.0861. According Matrix 1 2 3 4 5 6 7 8 9 10
dimension
to the consistency check, CI = 0.0103. Table 5 indicates that
RI = 0.90. RI 0 0 0.58 0.90 1.12 1.24 1.32 1.41 1.45 1.49

CI
CR = = 0.011 < 0.1. (4)
RI The judgment matrix is found to be valid. Then, the
individual fashion level in the classification of collocation is
The judgment matrix is found to be valid. Next, the solved with the following equation:
individual fashion level in the classification of make-up is
solved with the equation as follows: 5
𝐹2 = ∑𝜔2𝑗 𝑔𝑗 . (9)
4 𝑗=1
𝐹1 = ∑𝜔1𝑖 ℎ𝑖 . (5)
𝑖=1 (3) The criterion function of the weak feature is estab-
lished for weak feature in the classification of hair color.
(2) The criterion function of the weak feature is estab- If customers have the feature, the corresponding function
lished for weak feature in the classification of collocation. assignment of the feature is “1”; if not, the corresponding
If customers have the feature, the corresponding function function assignment is “0,” as indicated in the following
assignment of the feature is “1”; if not, the corresponding formula:
function assignment is “0”; namely,
{1, with the feature
{1, with the feature 𝑧𝑘 = { 𝑘 = 1, 2, . . . , 8. (10)
𝑔𝑗 = { 𝑗 = 1, 2, 3, 4, 5, 0, without the feature,
(6) {
0, without the feature,
{ By the means of expert advice and extensive survey,
In the combined form of expert advice and extensive matrix 𝐴 3 of feature in the classification of hair color is given
survey, matrix 𝐴 2 of the feature in the classification of make- as follows:
up is given as follows: 1 1 1 1 1
1 1 1
[ 2 2 4 3 4]
1 1 [ ]
1 2 2 [ 1 1 1 1 1]
[ 2 2 ] [1 1 1 ]
[ ] [ ]
[ 1 1 1 ] [ 2 2 4 3 4]
[ 1 1 ] [ ]
[ 2 3 3 ] [ 1 1 1 1 1]
[ ] [1 1 1 ]
𝐴2 = [ ]. (7) [ 2 2 4 3 4]
[ 1 1 1 ] [ ]
[ 1 1 ] [ 1 1 1]
[ 2 3 3 ] [ ]
[ ] 𝐴 3 = [2 2 2 1 1 ]. (11)
[ 1 1 ] [ 3 2 3]
[ 2 3 3 ] [ ]
[ 1 1 1]
[2 2 2 1 1 ]
[2 3 3 1 1 ] [ 3 2 3]
[ ]
[ ]
The weight of the weak feature is calculated based on [4 4 4 3 3 1 2 1]
[ ]
the feature matrix and the following results can be obtained: [ 1 1]
[3 3 3 2 2 1 ]
𝜔21 = 0.1194, 𝜔22 = 0.4053, 𝜔23 = 0.4053, and 𝜔24 = 0.0350. [ 2 2]
According to the consistency check, CI = 0.003325. RI = 1.12 [4 4 4 3 3 1 2 1]
by referring to Table 4.
The weight of the weak feature is calculated based on
CI the feature matrix, and the following results are obtained:
CR = = 0.002 < 0.1. (8)
RI 𝜔31 = 0.2731, 𝜔32 = 0.2731, 𝜔33 = 0.2731, 𝜔34 = 0.0701,
Scientific Programming 9

100.00 Table 6: SVM validation data.

Average accuracy (%)

95.00
Test frequency Accuracy
90.00
1 92.50%
85.00 2 90.00%
80.00 3 90.00%
75.00 4 97.50%
70.00 5 87.50%
0 10 20 30 40 50 60 70 80 90 100
Test times 6 92.50%
7 100.00%
Figure 11: SVM validation test.
8 87.50%
9 90.00%
𝜔35 = 0.0701, 𝜔36 = 0.0034, 𝜔37 = 0.0038, and 𝜔38 = 0.0034. 10 97.50%
According to the consistency check, CI = 0.008471. RI = 1.41 11 87.50%
by referring to Table 4. 12 97.50%
13 92.50%
CI
CR = = 0.006 < 0.1. (12) 14 95.00%
RI
15 82.50%
It can be found that the judgment matrix is valid. 16 82.50%
Subsequently, the individual fashion level in the classification 17 87.50%
of hair color is solved with the following equation:
18 95.00%
8 19 97.50%
𝐹3 = ∑ 𝜔3𝑘 𝑧𝑘 . (13) 20 85.00%
𝑘=1
21 90.00%
The overall fashion level is calculated according to the 22 87.50%
fashion level of each type of individual feature. Based on AHP, 23 90.00%
the weights of make-up, hair accessories, and hair color are 24 97.50%
relative to the weight of individual fashion levels: 𝜙1 : 𝜙2 : 𝜙3 = 25 97.50%
5 : 3 : 2.
26 92.50%
3 27 95.00%
𝐹 = ∑ 𝐹𝑚 𝜙𝑚 . (14) 28 92.50%
𝑚=1
29 92.50%
200 women are randomly selected as the validation object 30 100.00%
for model validation. Compared with the calculation results 31 92.50%
of the AHP model, the fashion level is obtained through 32 90.00%
experts scoring.
33 92.50%
34 85.00%
4. Analyses of Experimental Results 35 97.50%
(1) As shown in Table 6 and Figure 11, the average learning 36 90.00%
accuracy through the SVM classification experiment is as 37 97.50%
high as 92.08%. It suggests that the weak feature index is 38 90.00%
feasible and rational. However, there are three accuracies 39 85.00%
that are lower than 83% among the 200 validation groups; 40 95.00%
this is because of evaluation difficulties caused by unprecise
41 97.50%
image screening. In contrast to the two images with low
and high accuracies in the database, the research found that 42 95.00%
the database needs to be optimized. Thus, the weak feature 43 87.50%
shadow caused by the natural light should be adjusted. The 44 95.00%
dpi of image should be increased to 150%. In the future, the 45 100.00%
system data update should continue to be researched and 46 92.50%
analyzed.
47 92.50%
(2) The validation results through AHP model show that
48 92.50%
the accuracy of fashion level index reaches 92.37% and the
average matching degree of the classification results is 93%, 49 92.50%
10 Scientific Programming

Table 6: Continued. 80.00

70.00
Test frequency Accuracy 60.00
50.00
50 85.00% 40.00
30.00
51 92.50% 20.00
10.00

 (%)
52 85.00% 0.00
−10.00

1
33
65
97
129
161
193
225
257
289
321
353
385
417
449
481
513
545
577
609
641
673
705
737
769
801
833
865
897
929
961
993
53 90.00% −20.00
54 95.00% −30.00 Image number
−40.00
55 85.00% −50.00
−60.00
56 90.00% −70.00
−80.00
57 85.00%
58 95.00% Figure 12: AHP inaccuracy.
59 90.00%
60 97.50%
61 97.50% proving the effectiveness of AHP. Table 7 shows 20 groups of
62 97.50% results randomly; there are 16 values which are beyond 90%,
63 92.50% and 4 values are below 86%. The AHP experimental result, as
64 95.00% the verification method for SVM algorithm, indicates that the
65 92.50% SVM algorithm is effective. To further illustrate the results, 𝜁
66 82.50% is set as the inaccuracy; the equation is as follows:
67 95.00%
𝑉𝐶 − 𝑉0
68 92.50% 𝜁= . (15)
69 92.50% 𝑉0
70 82.50%
The change of 𝜁 is shown in Figure 12; fluctuation of 𝜁
71 95.00%
is in the range of plus or minus 15%. More than 700 pieces
72 92.50%
of data are fluctuated within 10%. It is certificated that the
73 90.00% AHP is effective. But there is a small amount of data with a
74 95.00% higher inaccuracy. Research found that it is mainly due to the
75 97.50% different psychological factors and physiological factors of the
76 87.50% ground truth. Also, there is some data quality which is below
77 95.00% standard. Later, it should be improved; on the whole, the
78 92.50% weak appearance features are suitable for the fashion clothing
79 90.00% recommendation.
80 87.50% The SVM classification model and AHP model are used to
81 97.50% explore the relationship between the appearance weak feature
82 95.00%
and fashion level. As can be seen from the results of the model
validation, both models can correctly characterize the fashion
83 95.00%
level, with the accuracy of being above 92%. Thus, it is a
84 92.50%
feasible scheme to classify the fashion level of different people
85 97.50% through the appearance weak feature, and the rationality of
86 90.00% weak feature index is proven.
87 92.50%
88 95.00%
89 87.50%
5. Conclusion
90 92.50% The paper uses the appearance weak feature to characterize
91 95.00% consumers’ fashion level and draws the following conclusions
92 97.50% by comparing the science experiment and expert evaluation:
93 97.50% the fashion level of consumers can be accurately determined
94 92.50% on the basis of make-up, accessories, and hair color. In the
95 87.50% paper, the support vector product is used to establish and
96 87.50% classify the fashion level model. According to the model
97 80.00% test results, the accuracy is more than 92%, and consumers’
fashion level is accurately classified. For verification from the
98 95.00%
perspective of the hierarchy by utilizing AHP, the accuracy
99 92.50%
is as high as 92.37%, demonstrating the effectiveness of
100 92.50% the appearance weak feature index. Customers can obtain
Average 92.08% the appropriate clothing recommendation based on their
Scientific Programming 11

Table 7: AHP data (choose 20 pieces of data as an example). [4] W. Zhang, B. Bo, M. Chu, J. Liu, and N. Yee, “Real-time
clothes comparison based on multi-view vision,” in Proceedings
Serial Calculation Data survey of the ACM/IEEE International Conference on Distributed Smart
Percentage
number results 𝑉𝐶 result 𝑉0 Cameras, pp. 1–10, Stanford, Calif, USA, September 2008.
1 0.6261 0.6933 89.26% [5] C. Limaksornkul, D. N. Nakorn, O. Rakmanee, and W. Viriy-
51 0.1691 0.1600 94.64% asitavat, “Smart closet: Statistical-based apparel recommenda-
101 0.6845 0.6500 94.97% tion system,” in Proceedings of the 3rd ICT International Senior
Project Conference (ICT-ISPC ’14), pp. 155–158, March 2014.
151 0.4484 0.4467 99.61%
[6] X. Hu, W. Zhu, and Q. Li, “HCRS, A hybrid clothes recom-
201 0.8061 0.7533 93.46% mender system based on user,” in Proceedings of the Ratings And
251 0.5903 0.6067 97.23% Product Features, pp. 270–274, 2014.
301 0.6772 0.7400 90.73% [7] L. Yu-Chu, Y. Kawakita, E. Suzuki, and H. Ichikawa, “Person-
351 0.4379 0.4267 97.43% alized clothing-recommendation system based on a modified
bayesian network,” in Proceedings of the IEEE/IPSJ 12th Interna-
401 0.6845 0.6367 93.02%
tional Symposium on Applications and the Internet (SAINT ’12),
451 0.6740 0.6700 99.41% vol. 59, pp. 414–417, July 2012.
501 0.5524 0.5567 99.22% [8] C. Limaksornkul, D. N. Nakorn, O. Rakmanee, and W. Viriy-
551 0.3949 0.4367 89.42% asitavat, “Smart closet: Statistical-based apparel recommenda-
601 0.6964 0.6700 96.20% tion system,” in Proceedings of the 2014 3rd ICT International
Senior Project Conference (ICT-ISPC ’14), pp. 155–158, March
651 0.6740 0.6900 97.62% 2014.
701 0.7956 0.7000 87.99% [9] S. Liu, T. V. Nguyen, J. Feng, M. Wang, and S. Yan, “Hi, magic
751 0.4553 0.4333 95.18% closet, tell me what to wear!,” ACM Multimedia, vol. 43, pp. 1333-
801 0.2978 0.3367 86.95% 1334, 2012.
851 0.7774 0.7033 90.47% [10] T. Iwata, S. Watanabe, and H. Sawada, “Fashion coordinates
recommender system using photographs from fashion maga-
901 0.6845 0.6800 99.35%
zines,” in Proceedings of the 22nd International Joint Conference
951 0.6811 0.6267 92.01% on Artificial Intelligence (IJCAI ’11), pp. 2262–2267, Catalonia,
Spain, July 2011.
[11] D. Sha, D. Wang, X. Zhou, S. Feng, Y. Zhang, and G. Yu,
classification results of the fashion level. By this means, fash- “An approach for clothing recommendation based on multiple
ion level evaluation method based on the appearance weak image attributes,” Web-Age Information Management, 2016.
feature can be successfully applied in the intelligent garment [12] L. C. Wang, X. Y. Zeng, L. Koehl, and Y. Chen, “Intelligent fash-
recommendation system, and it has essential significance for ion recommender system: Fuzzy logic in personalized garment
fashion recommendation. design,” IEEE Transactions on Human-Machine Systems, vol. 45,
no. 1, pp. 95–109, 2015.
[13] W. Hill, L. Stead, M. Rosenstein, and G. Furnas, “Recommend-
Conflicts of Interest ing and evaluating choices in a virtual community of use,” in
Proceedings of the Conference on Human Factors in Computing
The authors declare that they have no conflicts of interest. Systems, pp. 194–201, Denver, Colo, USA, May 1995.
[14] V. D. M. Ernani, E. A. Nogueira, and D. Guliato, “Content-Based
Acknowledgments Filtering Enhanced by Human Visual Attention Applied to
Clothing Recommendation,” in Proceedings of the International
The research is supported by the National Natural Science Conference on TOOLS with Artificial Intelligence, pp. 644–651,
Foundation of China (intelligent recommendation computa- 2015.
tion for electronic commerce based on eye fixation detection [15] S. Ajmani, H. Ghosh, A. Mallik, and S. Chaudhury, “An ontology
and emotion analysis, no. 61461021). based personalized garment recommendation system,” in Pro-
ceedings of the IEEE/WIC/ACM International Joint Conference
on Web Intelligence and Intelligent Agent Technology Workshops,
References WI-IATW 2013, pp. 17–20, November 2013.
[16] H. Li, F. Cai, and Z. Liao, “Content-Based Filtering Recommen-
[1] OECD, “Expanding the Garment Industry in the Kyrgyz Repub- dation Algorithm Using,” in Proceedings of the HMM, pp. 275–
lic,” OECD Eurasia Competitiveness Programme, 2014. 277, 2012.
[2] F. Cordier, H. Seo, and N. Magnenat-Thalmann, “Made-to- [17] E. A. Nogueira, E. V. De Melo, E. R. De Faria, and D. Guliato,
measure technologies for an online clothing store,” IEEE Com- “IKB-MS: A collaborative filtering approach associated with
puter Graphics and Applications, vol. 23, no. 1, pp. 38–48, 2003. human visual attention for clothing recommendation,” in Pro-
[3] R. Li, K. Zou, X. Xu, Y. Li, and Z. Li, “Research of interactive ceedings of the 21st Brazilian Symposium on Multimedia and the
3D virtual fitting room on web environment,” in Proceedings Web, WebMedia 2015, pp. 149–156, October 2015.
of the Research of Interactive 3D Virtual Fitting Room on Web [18] I. Hwang, M. Kim, and H. J. Ahn, “Data pipeline for generation
Environment. Symposium on Computational Intelligence and and recommendation of the IoT rules based on open text data,”
Design (Iscid ’11), vol. 1, pp. 32–35, Hangzhou, China, 2011. in Proceedings of the 30th IEEE International Conference on
12 Scientific Programming

Advanced Information Networking and Applications Workshops,

WAINA 2016, pp. 238–242, March 2016.
[19] M. Scholz, V. Dorner, M. Franz, and O. Hinz, “Measuring con-
sumers’ willingness to pay with utility-based recommendation
systems,” Decision Support Systems, vol. 72, pp. 60–71, 2015.
[20] Y. Freund, “Boosting a weak learning algorithm by majority,”
Information and Computation, vol. 121, no. 2, pp. 256–285, 1995.
[21] Y. Freund and R. Schapire, “A short introduction to boosting,”
Journal of Japanese Society for Artificial Intelligence, vol. 14, pp.
771–780, 1999.
[22] P. Viola and M. Jones, “Robust real-time face detection,”
Proceedings of the IEEE International Conference on Computer
Vision, vol. 2, p. 747, 2001.
[23] G. Tofighi, S. A. Monadjemi, and N. Ghasem-Aghaee, “Rapid
hand posture recognition using adaptive histogram template of
skin and hand edge contour,” in Proceedings of the 6th Iranian
Conference on Machine Vision and Image Processing (MVIP ’10),
October 2010.
[24] S. Soleimanizadeh, D. Mohamad, T. Saba, and A. Rehman,
“Recognition of Partially Occluded Objects Based on the Three
Different Color Spaces (RGB, YCbCr, HSV),” 3D Research, vol.
6, no. 3, article 22, 2015.
[25] Y. Zhang, “Search of snail recognition and counting based
on template matching of color and shape,” Journal of Gannan
Normal University, 2012.
[26] G. Rilling and P. Flandrin, “on the influence of sampling
on the empirical mode decomposition,” in Proceedings of the
IEEE International Conference on Acoustics Speed and Signal
Processing, pp. III-444–III-447, Toulouse, France, 2006.
[27] T. L. Saaty, “A scaling method for priorities in hierarchical
structures,” Journal of Mathematical Psychology, vol. 15, no. 3,
pp. 234–281, 1977.
 Advances in Journal of
Industrial Engineering
Multimedia
Applied
Computational
Intelligence and Soft
Computing
The Scientific International Journal of
Distributed
Hindawi Publishing Corporation
World Journal
Hindawi Publishing Corporation
Sensor Networks
Hindawi Publishing Corporation Hindawi Publishing Corporation Hindawi Publishing Corporation
http://www.hindawi.com Volume 2014 http://www.hindawi.com Volume 2014 http://www.hindawi.com Volume 2014 http://www.hindawi.com Volume 2014 http://www.hindawi.com Volume 201

Advances in

Fuzzy
Systems
Modelling &
Simulation
in Engineering
Hindawi Publishing Corporation
Hindawi Publishing Corporation Volume 2014 http://www.hindawi.com Volume 2014
http://www.hindawi.com

Submit your manuscripts at

-RXUQDORI
https://www.hindawi.com
&RPSXWHU1HWZRUNV
DQG&RPPXQLFDWLRQV  Advances in 
Artificial
Intelligence
+LQGDZL3XEOLVKLQJ&RUSRUDWLRQ
KWWSZZZKLQGDZLFRP 9ROXPH Hindawi Publishing Corporation
http://www.hindawi.com Volume 2014

International Journal of Advances in

Biomedical Imaging $UWLÀFLDO
1HXUDO6\VWHPV

International Journal of
Advances in Computer Games Advances in
Computer Engineering Technology Software Engineering
Hindawi Publishing Corporation Hindawi Publishing Corporation Hindawi Publishing Corporation Hindawi Publishing Corporation Hindawi Publishing Corporation
http://www.hindawi.com Volume 2014 http://www.hindawi.com Volume 2014 http://www.hindawi.com Volume 2014 http://www.hindawi.com Volume 201 http://www.hindawi.com Volume 2014

International Journal of
Reconfigurable
Computing

Advances in Computational Journal of

Journal of Human-Computer Intelligence and Electrical and Computer
Robotics
Hindawi Publishing Corporation
Interaction
Hindawi Publishing Corporation
Neuroscience
Hindawi Publishing Corporation Hindawi Publishing Corporation
Engineering
Hindawi Publishing Corporation
http://www.hindawi.com Volume 2014 http://www.hindawi.com Volume 2014 http://www.hindawi.com Volume 2014 http://www.hindawi.com Volume 2014 http://www.hindawi.com Volume 2014