Publication: Development of identity of interest (ioi) from sketch images in video surveillance system
Loading...
Date
2021-07-01
Authors
Tee, Han Shen
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
Face recognition system gains face recognition ability after learning each of the face features, providing a highly effective and accurate method in recognizing the identities of every person. Therefore, face recognition can be implemented in assisting the law enforcement where the police can search for suspect’s photo from their database quickly and effectively as well as identity tracking on suspects through video surveillance system can be done. However, in most cases, the photo image of a suspect is not available, therefore face recognition can only be performed on sketch drawing based on the recollection of eyewitnesses to identify the suspect. Photo-sketch recognition often yield less accurate results due to difference in texture between photo and sketch. There are 2 classical methods to improve the accuracy of photo-sketch recognition which are the generative method and discriminative method. In this project, a real-time sketch-based face recognition video surveillance system is developed where the system employs Haar-Cascade algorithm to detect and crop the face image appeared in camera, transformsthe face photo into face sketch through a generative method-based algorithm, GENRE model and lastly performs sketch-to-sketch recognition to determine the identity of face owner with VGG19 + SVM model which is trained with sketches. During the development of the system, several experiments to improve the performance of system on Demo images (probe images captured from demonstration video) have been conducted and the methods that have been determined which can maximize the system’s performance are the implementation of Segmented, GrayScaled and Sharpened Sketch-to-Segmented, GrayScaled and Sharpened Sketch image pre-processing technique and One Vs All SVM linear kernel, size increment of training dataset through addition of more sample images and application of face alignment on the involved images. As a result, the finalized system has recorded an identification accuracy of 88.3% on Demo images. Besides, the developed system is also tested on images of SCFace database to determine its performance in real case of video surveillance. The observation shows that the model is underperforming as it only obtains the best identification accuracy of 20.6% on the first 10 subjects in SCFace database although some improvements have been made. This indicates that it remains a challenge for the developed system to be implemented in video surveillance system.