EE604: Assignment 1

Prof. Tushar Sandhan

Due date: 15th Sept, 2023 Weight: 15%

Introduction
You need to figure out which methods and algorithms are suitable for solving below real-world
problems. Apart from this assignment, you are always welcome to practice basic image pro-
cessing algorithms by your own. Obviously, no one is stopping you for that. So, go forth and
pixelate, morph, and filter to your heart’s content.

1 Jai Hind: dronepositioner [5%]

(a) Input (b) Output for 1a (c) Input (d) Output for 1c

Figure 1: Two sample examples showing inputs and corresponding outputs.

The national flag holds significant cultural and symbolic importance for a nation. It serves as
a powerful emblem that embodies the history, values, unity of a country via evoking a sense of
patriotism and pride. Flags are more than mere pieces of cloth; they are visual representations
of a nation’s identity.
During aerial maneuvers, UAV or drones have to position themselves by controlling tilt and
height. Experimental robotics use ArUco markers (you can consider them as QR codes) which
are placed on the grounds for adjusting their position. However, fixed size visual patterns can be
used in remote terrain for drone positioning. Indian flag serves as a good localizer by military
drones. At the current non-optimal position, drone sees the image as shown in ‘input’, then it
adjusts its position (tilt and height) continuously with minimum movement till it sees the final
‘reference’ image. Note there is no need of rotation along Z-(height)-axis (yaw). Your task is to
generate the ‘reference’ image for the input seen by drone. Few examples are shown in Fig. 1.

1
 EE604:Assignment-1

Your function should take input as an image array and it should return the output as an
image array. Here input size may vary but each reference image is having a size 600x600 pixels,
radius of circle = 100, center of circle is (300,300), circle width = 2 and spoke width = 1 pixel. You
will be given the python program, where you have to complete the function while returning the
proper reference image for each input image. There will be 10 to 20 test cases used for automatic
evaluation of your program. Think about all corner cases.

2 eent ka jawab image se : brickquality [5%]

If you punch your hostel room’s wall in distress or in over-excitement, you might create a window
there. That would be a very high-quality punch; on the other side, there are poor-quality bricks
lying. There are four qualities of bricks used in the construction, having different costs according
to their quality. Sounds interesting? So they do, via making different sounds when struck
together. We will consider only 2 quality bricks here - low and high. Properly cooked red soil
bricks are of high quality. The furnace melts ferrites and imparts a metal-like quality to these
bricks. Therefore, when high-quality bricks are struck together, they produce a sound resembling
that of metal, while low-quality bricks sound like cardboard banged together.
You will be provided with audio (.mp3 files) of these brick sounds. You can analyze it in the
image domain via audio-to-spectrogram conversion using windowed Fourier transforms. You
may use various python libraries for getting spectrogram, like the below pseudo-code:

: import librosa
: y, sr = librosa.load(audio_path, sr=None) %sr=None preserves sampling rate
: n_fft = 2048 %FFT points, adjust as you need
: hop_length = 512 %Sliding amount for windowed FFT (adjust as needed)
: spec = librosa.feature.melspectrogram(y, sr, n_fft, hop_length, fmax=22000)
: # spec_db = perform power to decibel (db) image transformation of `spec'
: # decibel conversion → 10 log( maxP( P) ), where P is the spectral power
: # inspect spec_db images via displaying them before designing your algorithm

Write the python program, which can process the brick sounds and outputs their quality in
terms of metal (high) or cardboard (low) quality. Some representative sample inputs are given
in Fig. 2. Note your program will be tested for 10 to 20 different sample inputs, where program
should produce only one class output (either metal or cardboard) per sample testing.

(a) Input (b) Output for 2a (c) Input (d) Output for 2c

Figure 2: Sample inputs and outputs for both the classes ‘metal’ and ‘cardboard’. Your function
should take the audio input and it should return the class name as either ‘metal’ or ‘cardboard’.

2/3
 EE604:Assignment-1

3 Anuprasth Drishyam: horizontalviewer [5%]

Indian ancient Sanskrit literature holds a wealth of knowledge, encrypted in the language that
modern society is not fluent in. To digitize this knowledge, we need to scan the images and then
use optical character recognition (OCR), an image processing algorithms to digitize it. However,
many times, the scanned images are not horizontally aligned, which causes OCR to fail or pro-
duce erroneous digitization. These scanned images need to be re-rotated so that the final text
appears perfectly aligned horizontally in a readable form.
Design a python program to realign scanned text always in horizontal viewing manner. Fig. 3
shows some of inputs and corresponding output for these images. Your program will be tested
on various different input images where in the output any text character should not be cut and
image should be horizontally alinged. Output image size can be different than input image size.

(a) Input (b) Output for 3a (c) Input (d) Output for 3c

Figure 3: Sample inputs and outputs. Your function should take input as an image complete file
path (string) and it should write the output in the same working directory as .jpg image.

Submission
For each question use the original images that are provided to you separately apart from this
document. The only purpose of the figures here is to help with the proper explanations.
If your program does not produce output and throws errors, then you will receive a score
of 0 for that question. You alone are accountable for the submitted program’s proper operation.
Kindly make sure it executes and outputs just the things which are being asked for.
Depending on the output quality, partial to full credit will be awarded if the program executes
correctly. For plagiarized responses, even if your program does not execute or unexpectedly
produces correct answer, in all cases you will be awarded full marks with an honorarium prefix
as a negative sign.
Compress only the files requested into rollno.zip (eg.201234.zip) and upload it.

3/3