Foot Sole Based Personal Recognition

The most recent advances in security mechanisms have been based on technical biometric system approaches which use face,fingerprints,hand,veins,iris,and palms for purposes of identification and authentication.Foot sole identification offers a more effective personal authentication mechanism.It also has possible applications in law enforcement.To date,there is little statistical research regarding the uniqueness of the sole of each person's foot.Thus,the purpose of this thesis is to implement a foot sole based algorithm that can be used for personal recognition.In this thesis,I therefore propose a foot sole based personal recognition algorithm.What trails below is the brief explanation for the main aspects regarding my thesis.1)I collated foot sole data from 100 people.Ten images of each person's left foot sole were captured making a total of about 1000 foot soles images.2)I developed an image preprocessing algorithm.The acquired foot sole images are first changed to a black background and then cropped.The area of interest is obtained by resizing the foot sole images and then transforming into gray scale images.After which the image noise is removed from the selected area of interest.A process of adaptive histogram equalization was carried out for the purpose of minimizing and decreasing the appearance of false ridges in the foot sole images.After which the process of edge detection was also carried out using the canny detector and a double threshold was selected for each foot sole image.3)The ridges and bifurcations were being extracted from the forefoot part of the foot sole image using a cross number algorithm.Though there was some a bit of modification that was done during the process of marking and detecting the ridges and bifurcations.Subsequently,block regions of 41x41 size were extracted around the ridge ending.Then Histogram Oriented Gradients(HOG),Local Binary Patterns(LBP)and Speeded-Up Robust Features(SURF)characteristic features were extracted from the cropped block regions.4)I proposed a foot sole matching algorithm which was performed by calculating the distances between the block regions of size 41x41 around the ridges for the HOG,LBP and SURF feature vectors.The distances for the HOG,LBP and SURF features are computed between block regions for all the foot sole images in the dataset.The HOG,LBP and SURF distances were attained and summed up together to get the weighted distance.The minimum distances for the weighted distance from each of the block regions around the ridges were computed.The average distance being referred to as the weighted score for the minimum distances was calculated for each foot sole images.Sorting of the weighted score in ascending order with labels attached.Then KNN was used for classification.5)In terms of evaluating the accuracy and performance for the proposed algorithm,experimental analysis results show the recognition accuracy of 62%.