CV Project Objective of the Project: The human face has unique features which are useful to perform user authentication by face recognition. In this assignment, you will analyze the performance of a face biometric system considering two different feature binarization techniques. The first technique is “binarization by thresholding” which is covered in lecture. For the second one, you can choose any technique (along with feature pre-processing, optional though) but it should be able to improve the recognition performance than that of the first oneProblem summary and Data disclosure: In a face recognition system, a GallerySet is the database which contains the face templates of genuine users collected during enrollment phase whereas ProbeSet contains the face templates which are collected from the individuals who attempt to access the system as returning “users”. Two data files GallerySet.rar and ProbeSet.rar are provided for this project where the former contains 100 normalized face images from 100 individuals and the latter also contains 100 different images from the same 100 individuals. When you open the dataset, you will see the images are labelled with a subject number and an image number. The subject number denotes the individual’s ID and the image number denotes the sample number (‘1’ if it is gallery image and ‘2’ if it is probe image). The images who have the same subject number in the GallerySet and ProbeSet represent that they are the images from the same individual. For example: subject_img1 is the gallery image of an individual called “x” and subjectimg2 is the same subject’s probe image. For recognition purposes, you will mainly compare the probe images with the gallery images. To understand if the face recognition system is working or not, you may consider this criterion: “if a probe template demonstrates no to small distance with its corresponding gallery template but maintain high distance with other gallery templates, then you may consider that the face recognition system is functioning.” In this part, you will build a face recognition system called “System_A” which will consider binary features generated by “binarization by thresholding” technique. Complete the following steps to design System_A: (a) Both the GallerySet and ProbeSet contain gray-toned 3 channel images. Read them as single channel grayscale image and perform image binarization on them by “binarization by thresholding” method which was covered in lectures. Consider a threshold of 128 where if a pixel’s intensity is <=128, it will be “0” and if the intensity is > 128 then it will be “1” in the binary image. You will get binary features for all the gallery and probe templates after this process. Now considering the Hamming Distance function as your distance measure, perform the comparison between each binary probe template with all the binary gallery templates and generate a score matrix. Your score matrix A will be a 100x100 matrix where A[i,j] denotes the distance score generated by comparing the i-th probe with the j-th gallery data. Provide a snippet of the A[0:9,0:9] matrix. There are several performance metrics available to evaluate the performance of a face recognition system. A decidability index value (d’) is one of them which will be considered for this project. It quantifies the separation between the distribution of genuine scores and imposter scores. A score is called genuine if it is produced by comparing a probe template with a gallery template where both the templates belong to the same individual. On the other hand, an imposter score is produced when a probe template was compared to a gallery template where the templates belong to two different individuals. Using the genuine and imposter scores from the score matrix from previous step, determine the d’ using the given formula. (10 points) Note: μ1 and μ0 are the mean of the genuine and imposter score distribution respectively. Similarly, σ1 and σ0 denote the standard deviation of the genuine and impostor distribution respectively.Consider a face recognition system named “FaceRec” which has a decidability index of 3.5. Do you think your system will have a better performance than that of FaceRec? Why or why not? Justify your answer. Part I represents a very basic face biometric system which considered binary features generated from the traditional binarization by thresholding technique. In part II, you are going to build another face recognition system called System_B where you will consider a different feature binarization technique so that the performance is improved from that of System_A. You are allowed to include any pre-processing or post-processing steps and/or binary feature comparison technique of your choice as well for this part. By adopting any image/feature binarization technique of your choice, generate new binary features for all the images from GallerySet and ProbeSet. You are allowed to include any pre-processing or post-processing steps if you want, but the features have to be still in binary format. Now considering any distance measure function (must be compatible with binary features), perform the comparison between each binary probe template with all the binary gallery templates to generate a score matrix. Your score matrix B will also be a 100x100 matrix where B[i,j] denotes the distance score generated by comparing the i-th probe with the j-th gallery data. Provide a snippet of the B[0:9,0:9] What is the decidability index for System_B? The goal of Part II is to construct System_B in such a way so that it can provide better recognition performance than that of System_A. Are you able to ensure the improvement? Report the improvement factor (IF) considering the given formula. (20 points) IF = round (delta d’, 2) where delta_d’ = d’ of System_B – d’ of System_A Points breakdown for this question: Improvement_factor (IF) Points you may earn -ve 0 0 5 ≥ 0.1 but < 0.4 10 ≥ 0.4 but < 0.9 15 ≥ 0.9 but < 1.5 18 ≥ 1.5 20 (