Designing an Illumination Effect Canceling Filter in Facial Images for Multi-View Face Detection and Recognition in Images with Complex Background Reza Shoja Ghiass, Emad Fatemizadeh, Member, IEEE, Farrokh Marvasti, Senior Member, IEEE Department of Electrical Engineering, Sharif University of Technology, Tehran, Iran. Emails: reza.shoja@gmail.com, fatemizadeh@sharif.edu, marvasti@sharif.edu Abstract- This paper presents a novel approach for detection and recognition of multi-view faces whose location is unknown and the illumination conditions are varying. The illumination is a big problem in the face detection and recognition aspects. Our proposed method doesn’t use the skin color information for face detection. The detection of faces is accomplished after canceling the effect of the various illumination conditions. Two completely different methods are proposed for face detection in this paper. Because of the independency of the approaches to the face’s skin color, the persons with every kind of skin colors are detected even in completely dark environments. Next, the detected faces are recognized. The experiments have shown that we can combine some proposed aspects of the feature based methods with the eigenface method and get very successful results. The illumination dependency problem of the eigenface method has also been solved by a new methodology in this paper. I. INTRODUCTION Edge information is a significant object representation feature that is insensitive to illumination changes, but to certain extent. There are a few works on face recognition aspect using edge of faces [1]. Poggio [2] reported that to have a successful object recognition approach, one might need to combine the aspects of feature-based approaches with template matching method. In our work, we will show a successful face recognition method that doesn’t combine the aspects of feature-based approaches with template matching method. The new work combines some proposed edge information with the eigenface method [3]. Eigenface method has been proved to be very successful in face recognition. This method is the implementation of Principle Component Analysis (PCA) in face recognition. Kirby and Sirovich [4] reported that illumination normalization is necessary for the eigenface method. In this paper, we have also solved the illumination dependency problem of eigenface approach by a new method instead of illumination normalization. In most of the researches about faces, there are some available face databases that the faces available in these databases have known size and position. Therefore, in practical aspects, the faces have to be detected from images before face recognition. There are many face detectors that detect faces using the information of the face’s skin color. The great problem with the skin color based face detectors is that they will not satisfy the results, if the illumination conditions are changed intensely, and the persons with different kind of skin colors will not be detected under varying illumination conditions. These problems are solved in our proposed methods. The outline of this paper is as follows: Our approach to multi-view face detection and recognition under varying illumination conditions is described in Section II. Section III provides experimental results, while the conclusions are given in Section IV. II. MULTI-VIEW FACE DETECTION AND RECOGNITION SYSTEM In this section, a novel method for multi-view face detection and recognition under varying illumination conditions is introduced. As we know, the feature based face recognition methods have robust performance under different lighting conditions. On the other hand, the eigenface algorithm is proved to be very successful in face recognition. This method uses the face pictures as raw data and extracts some feature vectors from these pictures by the PCA algorithm. So the classifier can be trained with these feature vectors. If the illumination conditions of the environment changes, then the efficiency of the system will decrease because the value of the pixels in the test image will change significantly. This is a big problem with the eigenface method. If the edge information of face is applied in eigenface method, instead of the face image, we anticipate that the illumination dependency problem of eigenface method will be resolved. This means that we want to extract eigenvectors from edge information of faces and not from face images. The advantage of this idea, if can be implemented, is the robustness of eigenface method under varying illumination conditions. For this reason, we use the various types of edge detection filters on the UMIST [5] database to find out which of these filters has better performance. Thus, we extract the edge information of the faces available in the database and implement the eigenface method on the edge information of faces to extract some 2008 Internatioal Symposium on Telecommunications 978-1-4244-2751-2/08/$25.00 ©2008 IEEE 809