Scribd
Upload
64 views

Face Identification Using Large Feature Sets: Abstract

The document proposes a face identification method using large feature sets that is scalable to gallery size. It uses a tree-based structure where at each node, a binary classifier based on PLS regression guides the search for a matching subject. This provides substantial reduction in comparisons compared to one-against-all approaches. Existing methods are statistical learning or local matching, but the proposed method extracts HOG, LBP, color, and Gabor features from image blocks, applies PLS dimension reduction, and uses a classification-regression tree for supervised learning and verification.

Uploaded by

Soujanya Reddy
Copyright
© Attribution Non-Commercial (BY-NC)
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as DOC, PDF, TXT or read online on Scribd
64 views

Face Identification Using Large Feature Sets: Abstract

The document proposes a face identification method using large feature sets that is scalable to gallery size. It uses a tree-based structure where at each node, a binary classifier based on PLS regression guides the search for a matching subject. This provides substantial reduction in comparisons compared to one-against-all approaches. Existing methods are statistical learning or local matching, but the proposed method extracts HOG, LBP, color, and Gabor features from image blocks, applies PLS dimension reduction, and uses a classification-regression tree for supervised learning and verification.

Uploaded by

Soujanya Reddy
Copyright
© Attribution Non-Commercial (BY-NC)
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as DOC, PDF, TXT or read online on Scribd
You are on page 1/ 5

Face Identification Using Large Feature Sets

Abstract: We propose a method which is scalable to the gallery size and modify one-against-all approach to use a tree based structure. At each internal node of the tree , a binary classifier based on PLS regression is used to guide the search of the matching subject in the gallery. The use of this structure provides substantial reduction in the number of comparisons when a probe sample is matched against the gallery. There are very accurate techniques to perform face identification in controlled environments, particularly when large numbers of samples are available for each face. However, face identification under uncontrolled environments or with a lack of training data is still an unsolved problem. Existing System Face recognition research has achieved significant progress over the past decade. There are some recent and comprehensive surveys on face recognition written by Tolba et al. and Zhao et al. Most previous works are based on statistical learning or local matching methods . Methods in the former category use the whole face region to perform recognition and include techniques such as subspace discriminant analysis, support vector machine (SVM), and AdaBoost. These methods suffer from the generalizability problem due to the unpredictable distribution of real-

world testing face images, which might be dramatically different from the training samples . Methods in the latter category first locate several facial features and then classify faces according to local statistics. The use of low-level feature descriptors has been an effective approach in face recognition. Scale-invariant feature transform (SIFT) and histogram of oriented gradients (HOG), which can be viewed as a quantized code of the facial gradients, are used in face recognition as effective descriptors. Local binary patterns (LBP) and Gabor filters are descriptors most widely used in face recognition. LBP is invariant to monotonic photometric change and can be efficiently extracted. Gabor features are characterized by spatial frequency, spatial locality, and orientational selectivity for coping with image variability such as illumination variations. There are several combinations or variations based on these LBP and Gabor descriptors . In addition, by varying a sampling radius and combining the LBP images, a multiresolution representation based on LBP called multiscale LBP (MSLBP) can be obtained. This representation has been suggested for texture classification, and the results reported for this application show better accuracy than that of the single-scale LBP method. Recent research has focused on parameter learning with a HOG-like template . Other LBP variants, including three-patch-based LBPs, four-patch-based LBPs , and Patterns of Oriented Edge Magnitudes have been introduced for face recognition.

Proposed System Our proposal begins by decomposing the images into nonoverlapping blocks. Each block is subjected to feature extraction namely, Shape Features HOG Texture Features LBP Color Features Averaging the intensities in each block Visual Features Gabor Features These features are subjected to dimension reduction using PLS regression. Finally, these features are labeled and learnt by Tree based structure called as Classification Regression Tree. Using this supervised learning, the verification is carried out for a probe sample. Identification accuracy is calculated and analyzed for justification.

Modules Image Enhancement Feature Extraction HOG features LBP features Colour features Gabor features PLS model Classification Tree Dataset Classification Software Specification: Hardware Requirement: Pentium IV 2.7 GHz 1GB DDR RAM

250Gb Hard Disk Software Requirement: Operating System : Windows XP Tool : Matlab

Version : 7.9

You might also like