Face Recognition And Matching Enhancement Based On Wavelet Transform

Face recognition is an important topic in the field of image processing and computer vision. The aim of face recognition is to detect which face class the input test face image belongs to, by matching it to large database. The face recognition rate refers to the ratio of the correct recognized face images. This rate can be enhanced either by enhancing the face image itself using image processing principles or by enhancing the techniques used for recognition. In this dissertation both enhancement methods are used:wavelet image denoising filters are used for preprocessing the face image and a novel face recognition technique is proposed based on principal components analysis, then human identification approach is proposed to enhance the matching process based on scale invariant features transform (SIFT). The main research contents in this dissertation are outlined as follows:1. A comparison between face recognition rate with noise and face recognition rate without noise is produced. We applied the wavelet based image denoising filters on ORL database to create new databases, and then the face recognition rate is calculated to them. Image denoising using single filter (Haar, Daubechies and Symlet), image file formats (JPG and BMP) and two pre-processing chains are proposed to enhance the recognition rate of some face recognition techniques. For enhancement by single filter three experiments are conducted. In the first experiment, different wavelet filters with different levels of decomposition (up to ten decompositions) are used for denoising the ORL database and the comparison is done when Principal Components Analysis (PCA) is applied to evaluate the verification rate. In the second experiment, denoising different sets of ORL database with methods that have the best performance in levels (1,2,3, and10) is done (as a result from experiment1). In the third experiment, the proposed HaarlO method was implemented on the face recognition approaches such as PCA, Linear Discriminate Analysis (LDA), Kernel PCA and Fisher Analysis (FA), and the recognition rates are evaluated for both the noisy and de-noisy databases. For enhancement based file JPG and BMP face image file formats, the effect of file formats with and without wavelet denoising process are studied in this dissertation to evaluate the performance of the same face recognition techniques used for single filter. Also, two adaptive image enhancement and denoising chains are introduced in this dissertation. Each processing chain consists of three steps. The first chain is proposed to enhance Principal Component Analysis (PCA) and Kernel Principal Component Analysis (KPCA) recognition rate. In the first step of this chain, the face images are denoised with Haar wavelet denoising filter at level ten of decomposition, in the second step, the denoised image is adjusted to enhance the image contrast, and in the third step the high pass filter ’Laplacian of Gaussian filter’ is used for detecting edges in face images. The second chain is proposed to enhance Linear Discriminate Analysis (LDA) and Kernel Fisher Analysis (KFA) recognition rate. In the first step of this chain the image contrast is adjusted, histogram equalization is carried out as second step, and in the third step the image is de-noised with Haar wavelet denoising filter at level ten of decomposition.Our proposed enhancement approaches produced superior results and raised the recognition rate in PCA, LDA, KPCA, and FA up to10%,5%,20%and4%respectively when400face images are used.2. A novel face recognition approach based on Adaptive Principal Component Analysis (APCA) and de-noised database is presented. The aim of our approach is to overcome PCA disadvantages, especially the two limitations of discriminatory power poverty and the computational load complexity, by producing a new adaptive PCA based on single level2-D discrete wavelet transform using Daubechies filter mode. All face images in ORL database are transformed to JPG file format and are de-noised by Haar wavelet at level10of decomposition, which is to exhibit the advantage of wavelet over compressed JPG files instead of using the original PGM file format. As a result, our adaptive approach produced good performance in raising the accuracy ratio and reducing both the time and the computation complexities when compared with four other methods represented by standard statistical PCA, Kernel PCA, Gabor PCA and PCA with Back propagation Neural Network (BPNN).3. Two novel human face identification approaches are presented as two parts. Part1describes the first approach based on adaptive PCA when applied on non denoised face database. Part2describes the second approach based on adaptive APCA when the face image denoised with double wavelet filter Bior1.1-Haar at level ten of decomposition. The both adaptive approaches entered to SIFT feature extraction method to generate two adaptive PCA-SIFT matching approaches in wavelet domain rather than the mathematical computation and representation of PCA that entered to SIFT, our matching approaches is evaluated separately and compared with the standard PCA-SIFT matching approach.