International Journal of Science and Research (IJSR) ISSN (Online): 2319-7064 Impact Factor (2012): 3.358 Volume 3 Issue 9, September 2014 www.ijsr.net Licensed Under Creative Commons Attribution CC BY Multiscale Edge Based Approach for face Identification Suvarna Joshi 1 , Dr. Abhay Kumar 2 1 School of Electronics, Devi Ahilya University, Indore, India Abstract: A novel and uniform edge based framework for both face identification and verification is presented in this paper. This paper proposes the feature extraction algorithm based on different directional information and multiscale information of face. Proposed feature extraction algorithm extract discrete wavelet transform edge features to determine facial features. DWT acts as a main contributor in efficient feature extraction which results in the high recognition rates. After generating feature vectors, distance classifier is used for feature classification step. We have evaluated performance of proposed multiscale edge based method on standard databases with variation in pose, expression, orientation, lightning, facial information.ORL, JAFFE and IIT face databases were used to generate the results. Experiments show that edge wavelet features can significantly improve system performance.DWT and edge detection based approach also causes reduction in feature vector size. Thus, the proposed method has proven to be a promising technique under arbitrary variations in illumination, poses and backgrounds with slight occlusions too. Keywords: multiscale, wavelet transform, edge tracking 1. Introduction Face recognition is becoming an emerging technology from last few decades.[4][5] Biometric recognition describes a unique model as it is doesn't need any knowledge from the subject for its use. There is lot of increase in applications based on circuit television (CCTV) and other forms of surveillance which facilitates smart methods to effectively identify and validate persons within a clip. Some applications can also require to identify features of the subjects, gender, ethnicity, age, etc. The Designing of algorithm for face recognition system is a sophisticated problem because faces are having lot of variations and may be located in a different changed environment. Face recognition is a challenging problem due to the wide variety of illumination, facial expression and pose variations, lightning conditions, ageing, disguises such as slight cut, glasses or makeup. These problems cause degradation in performance of face recognition system. Development of face recognition system involves selection of suitable features to represent a face under environmental changes. This paper proposes a new approach based on edge detection algorithm. Directional information is calculated using Fast Wavelet Transform. Whole face information is generated using Discrete Wavelet Transform for robustness of multi resolution and multi directional approach for pose and expression invariant face recognition. Fusion is done using approximate sub-bands at each level of discrete wavelet transform. In order to capture better multiscale information in face images, each face image is described by a subset of multi-scale wavelet coefficients obtained from edge detected images which characterize the face textures. The overall texture features of an image at each resolution are obtained through fusion of the detailed sub bands. We fused sub band coefficient from which we extract compact and meaningful feature vectors using statistical measures such as feature variance. 2. Related Work Several authors have carried out their research work in face recognition. There have been many methodologies proposed in the literature for solving the problem associated with face recognition. Some of methods describes allows feature presentation in terms of principal components analysis [2], discrete wavelet transform [10][13]. Gabor wavelet-based face feature representation provides an excellent methodology for design of robust face recognition system. For this reason, Gabor wavelets are most commonly used approach for development of face recognition system [6][11][8][12].Several wavelet transform methods have been proposed in the recent years to address the problem of face recognition [14][15][7]. Gabor wavelet-based face image representation approach is an efficient approach. Still it has two important drawbacks, it is computationally very complex and multi-scale multidirectional approach results in increase in processing time and also memory requirements. For an input image of size 128x128 pixels, size of the Gabor feature vector will be equal to 655360 pixels for 8 directions and 5 scales representation. Wavelet based multi-resolution approach is useful to overcome effects non-uniform illumination [9].In this paper, we have used a unique combination of edge detection and wavelet transform based algorithm to achieve superior feature extraction which greatly contributes to higher recognition rate 3. Proposed Method This paper presents a novel feature extraction algorithm based on wavelet coefficients extracted from edge tracked images. Edge tracking method allows for generation of an image in which edges are highlighted from background image. An edge is a set of pixels in which the image intensity rapidly changes. The edge detection process allows simplifying the analysis of images by drastically reducing the amount of data to be processed, while at the same time preserving useful structural information about object boundaries Edge detection allows for preservation of structural properties of an image and results in significant reduction of amount of data. Proposed methodology for face feature extraction aims at extraction structural information from face image using canny edge detection algorithm Paper ID: SEP14573 2427