Feature selection (FS) is a global optimization problem in machine learning, which reduces the number of features, removes irrelevant, noisy and redundant data, and results in acceptable recognition accuracy. It is the most important step that affects the performance of a pattern recognition system. This paper presents a novel feature selection algorithm based on particle swarm optimization (PSO). PSO is a computational paradigm based on the idea of collaborative behavior inspired by the social behavior of bird flocking or fish schooling. The algorithm is applied to coefficients extracted by two feature extraction techniques: the discrete cosine transforms (DCT) and the discrete wavelet transform (DWT). The proposed PSO-based feature selection algorithm is utilized to search the feature space for the optimal feature subset where features are carefully selected according to a well defined discrimination criterion. Evolution is driven by a fitness function defined in terms of maximizing the class separation (scatter index). The classifier performance and the length of selected feature vector are considered for performance evaluation using the ORL face database. Experimental results show that the PSO-based feature selection algorithm was found to generate excellent recognition results with the minimal set of selected features.

Face recognition using particle swarm optimization-based selected features

This paper presents two new techniques, viz., DWT Dual-subband Frequency-domain Feature Extraction (DDFFE) and Threshold-Based Binary Particle Swarm Optimization (ThBPSO) feature selection, to improve the performance of a face recognition system. DDFFE uses a unique combination of DWT, DFT, and DCT, and is used for efficient extraction of pose, translation and illumination invariant features. The DWT stage selectively utilizes the approximation coefficients along with the horizontal detail coefficients of the 2-dimensional DWT of a face image, whilst retaining the spatial correlation of pixels. The translation variance problem of the DWT is compensated in the following DFT stage, which also exploits the frequency characteristics of the image. Then, all the low frequency components present at the center of the DFT spectrum are extracted by drawing a quadruple ellipse mask around the spectrum center. Finally, DCT is used to lay the ground for BPSO based feature selection. The second proposed technique, ThBPSO, is a novel feature selection algorithm, based on the recurrence of selected features, and is used to search the feature space to obtain a feature subset for recognition. Experimental results obtained by applying the proposed algorithm on seven benchmark databases, namely, Cambridge ORL, UMIST, Extended Yale B, CMUPIE, Color FERET, FEI, and HP, show that the proposed system outperforms other FR systems. A significant increase in the recognition rate and a substantial reduction in the number of features required for recognition are observed. The experimental results indicate that the minimum feature reduction obtained is 98.2% for all seven databases.

Face recognition using transform domain feature extraction and PSO-based feature selection

This paper presents a face recognition method based on Discrete Cosine Transform (DCT) coefficient's selection. Without a normalization phase, the proposed method uses, in its feature selection stage, a technique based only on the DCT coefficients amplitudes. Three coefficient selection criterions were analyzed: the first one is the average of the coefficients' amplitudes; the second one is based on counting the occurrence of each coefficient, which are stored in a set of lists containing the most significant coefficients; finally, the third criterion is based on the average position of the coefficients in a list of coefficients ordered by amplitude. Experimental tests on the ORL Face Database [1] achieved 99.00% of recognition accuracy using only 50 DCT coefficients, with low computational cost. Additionally, the method achieved 100.00% of recognition accuracy when the correct face is within a range of eight returned faces.

Face recognition using DCT coefficients selection

Feature selection (FS) is a global optimization
problem in machine learning, which reduces the number of
features, removes irrelevant, noisy and redundant data, and
results in acceptable recognition accuracy. This paper presents a
novel feature selection algorithm based on Bacteria Foraging Optimization (BFO). The algorithm is applied to coefficients extracted by discrete cosine transforms (DCT). Evolution is
driven by a fitness function defined in terms of maximizing the class separation (scatter index). Performance is evaluated using the ORL face database.

https://thesai.org/Downloads/SpecialIssueNo3/Paper 17-Face Recognition Using Bacteria Foraging Optimization-Based Selected Features.pdf

Face Recognition Using Bacteria Foraging Optimization-Based Selected Features

Face recognition under varying occlusions and lighting conditions is challenging, and exacting occlusion and illumination invariant features is an effective approach to solve this problem. In this paper, we propose two novel techniques viz., DWT Thresholding and Laplacian-Gradient Masking, to improve the performance of a face recognition system. DWT Thresholding is used to extract the approximation coefficients along with the prominent detail coefficients of the 1-dimensional DWT of an image, thereby selecting only relevant features and enhancing face recognition. Laplacian-Gradient Masking is a pre-processing technique which combines the edge detection properties of both the Laplacian and the Gradient operators to create a masked image, containing a well-defined contour of the prominent facial features. The resulting pre-processed image contains the salient edge details of the face and prepares the ground for feature extraction. Experimental results show the promising performance of DWT Thresholding and Laplacian-Gradient Masking for face recognition on ORL, UMIST, Extended Yale B and Color FERET databases.

Face recognition using DWT thresholding based feature extraction with laplacian-gradient masking as a pre-processing technique

This paper describes two methods for face recognition problem with an image database (DB) The two methods are based on face-based approach The first method combines the wavelet transform (WT) and fast Fourier transform (FFT), while the second method, combine the WT and discrete cosine transform (DCT) Using the virtue of WT we extract the most insensitive features to facial expression variations The first method is proven to be very efficient with face images of different features, illuminations and small occlusion The second method has proven to be good with the above variations of first method together with face images of different: scales, poses and rotated images (plusmn20) Two DBs are used to perform our work The Yale DB and Olivetti DB The two introduced methods are compared to two known methods, the template method and the eigenface method on the same DBs

S. El Taweel Gad Allah

Papers

Face recognition using wavelet transform, fast Fourier transform and discrete cosine transform