Optimizing Face Recognition in Images

The paper aims to optimize practical applications for detection and face recognition using HaarLike classifiers in conjunction with a centroid algorithm to detect the gaze direction. For selecting method we have implemented principal components analysis (PCA), and for image resizing after detected we used the interpolation method. The proposed algorithm is tested by a database of its own and by training aiming at detecting and recognizing faces in crowds crowded. Practical application in Microsoft Visual Studio was held in CSharp using predefined elements in Open Cv. Finally we present comparative results with other three systems that implement biometric technologies, about false acceptance rate, false rejection rate, and processing time. Keydords: recognition, image, technologies, processing, detection Optimizing Face Recognition in Images 124 Essentially, the method consists in transforming space training vectors into a new space whose dimension is equal to the number of classes. Such vectors are transformed pseudofeatures are subsequently presented a classical multilayer networks. Finally, it must be made and natural observation that because of fast development and implementation of new structures of neural networks, neural computation range theory applied in pattern recognition (visual) is much larger, far exceeding the possibilities of coverage and presentation in an article. 3. CUBIC INTERPOLATION Resizizing and image detection for forced comparing the same size as the test image was performed using cubic interpolation methodology. This approach assumed predefined approximation techniques based on Spline function. It was proposed that computer application, after applying interpolation to achieve: -Display the added image to a grayscale Save a text file of faces involved writing the labels of involved faces in a text file for loading and subsequent detection Interpolation is a method of estimating the values applied in a location without measurements, based on measured values in neighboring points. The process consists in finding a function f (x, y) to represent the entire surface z values associated with points (x, y) arranged regularly performing a prediction function z values for other positions arranged regularly. The considerations that led to the choice of interpolation were offering a large space for data input processing, a very short time and the possibility of implementing applications using open source. Implementation method of spline functions in the process of cubic interpolation. For the development of facial recognition applications to people insisted that uses triangulation method. In reference [2] is demonstrated that PCA allows to describe variations between images of models with significant differences in features. In reference [3] it was shown that using 16 subjects the three types of image formats by changing lighting conditions that the 6 types of resolution (512 x 16 x 16 ... 512), for a total of 2592 images entrained can get a correct recognition rate of 96%, above the normal lighting of the light 85% below to 64% above the size of the image. The results showed that the maximum 19% can be obtained by changing the lighting rejections, 39% and 60% by changing the orientation by changing the size of the acquired images. All these considerations led us to the decision to use the selection method using PCA in practical applications performed in this paper. 2. THE CLASSIFICATION ITSELF Implementation of neural structures as forms of visual classifiers is one of the most common applications of neural networks. The training of a neural network for visual pattern recognition (2D or 3D) requires, in principle, the approach has three distinct directions, generic schematic below: (1) {visual forms (2D or 3D)} entry extraction / selection shape descriptors (1D) {lot of training (1D)} standard neural network; (2) {visual forms (2D)} input {lot of training (2D)} specialized 2D neural networks; {Visual forms (3D)} entry extraction / selection projections (2D) {lot of training (2D) 2D specialized neural networks; (3) {visual forms (3D)} entry extraction / selection shape descriptors (3D) {lot of training (3D)} 3D specialized neural networks. Neural networks listed in the previous paragraph (2) are specialized structures involvement with 2D input forms, the flexibility organization of neurons in the input layer in the form of two-dimensional arrays of different shapes (circular, hexagonal, etc.). An efficient method which eliminates the step neural classification of feature extraction is disclosed in reference [4].