Hazem M. El-Bakry

Face detection using fast neural networks and image decomposition

Abstract In this paper, a new approach to reduce the computation time taken by fast neural nets for the searching process is presented. The principle of divide and conquer strategy is applied through image decomposition. Each image is divided into small in size sub-images and then each one is tested separately using a fast neural network. Compared to conventional and fast neural networks, experimental results show that a speed up ratio is achieved when applying this technique to locate human faces automatically in cluttered scenes. Furthermore, faster face detection is obtained by using parallel processing techniques to test the resulting sub-images at the same time using the same number of fast neural networks. Moreover, the problem of sub-image centering and normalization in the Fourier space is solved.

Fast pattern detection using normalized neural networks and cross-correlation in the frequency domain

Neural networks have shown good results for detection of a certain pattern in a given image. In our previous work, a fast algorithm for object/face detection was presented. Such algorithm was designed based on cross-correlation in the frequency domain between the input image and the weights of neural networks. Our previous work also solved the problem of local subimage normalization in the frequency domain. In this paper, the effect of image normalization on the speedup ratio of pattern detection is presented. Simulation results show that local subimage normalization through weight normalization is faster than subimage normalization in the spatial domain. Moreover, the overall speedup ratio of the detection process is increased as the normalization of weights is done offline.

New faster normalized neural networks for sub-matrix detection using cross correlation in the frequency domain and matrix decomposition

Neural networks have shown good results for detecting a certain pattern in a given image. In this paper, faster neural networks for pattern detection are presented. Such processors are designed based on cross correlation in the frequency domain between the input matrix and the input weights of neural networks. This approach is developed to reduce the computation steps required by these faster neural networks for the detection process. The principle of divide and conquer strategy is applied through matrix decomposition. Each matrix is divided into smaller in size submatrices and then each one is tested separately by using a single faster neural processor. Furthermore, faster pattern detection is obtained by using parallel processing techniques to test the resulting submatrices at the same time using the same number of faster neural networks. In contrast to faster neural networks, the speed up ratio is increased with the size of the input matrix when using faster neural networks and matrix decomposition. Moreover, the problem of local submatrix normalization in the frequency domain is solved. The effect of matrix normalization on the speed up ratio of pattern detection is discussed. Simulation results show that local submatrix normalization through weight normalization is faster than submatrix normalization in the spatial domain. The overall speed up ratio of the detection process is increased as the normalization of weights is done off line.

Comments on using MLP and FFT for fast object/face detection

A fast neural nets for object/face detection are presented in [(S.Ben-Yacoub, 1997), (Beat Fasel, 1998) and (S. Ben-Yacoub et al., 1999)]. The speed up factor of these networks based on cross correlation in frequency domain between the input image and the weight of the hidden layer. But, these equations presented in [(S. Ben-Yacoub, 1997), (Beat Fasel, 1998) and (S. Ben-Yacoub et al., 1999)] for conventional and fast neural nets as well as speed up ratio are not valid for many reasons presented here. In this paper, a correct formula for the computation steps required for conventional, fast neural nets presented in [(S. Ben-Yacoub, 1997), (Beat Fasel, 1998) and (S. Ben-Yacoub et al., 1999)] and speed up ratio is introduced. Moreover, conventional neural nets are proved to be faster than those fast neural nets presented in [(S.Ben-Yacoub, 1997), (Beat Fasel, 1998) and (S. Ben-Yacoub et al., 1999)]. Practically, simulation results confirm this approval. Furthermore, only in case that the input image is symmetric or the weight are symmetric, neural nets presented in [(S. Ben-Yacoub, 1997), (Beat Fasel, 1998) and (S. Ben-Yacoub et al., 1999)] give correct result as conventional neural nets.

New fast time delay neural networks using cross correlation performed in the frequency domain

This paper presents a new approach to speed up the operation of time delay neural networks. The entire data are collected together in a long vector and then tested as a one input pattern. The proposed fast time delay neural networks (FTDNNs) use cross correlation in the frequency domain between the tested data and the input weights of neural networks. It is proved mathematically and practically that the number of computation steps required for the presented time delay neural networks is less than that needed by conventional time delay neural networks (CTDNNs). Simulation results using MATLAB confirm the theoretical computations.

Face detection using neural networks and image decomposition

An approach to reducing the computation time taken by fast neural nets for the searching process is presented. The principle of the divide and conquer strategy is applied through image decomposition. Each image is divided into small in size sub-images and then each one is tested separately using a fast neural network Compared to conventional and fast neural networks, experimental results show that a speed up ratio is achieved when applying this technique to locate human faces automatically in cluttered scenes. Furthermore, faster face detection is obtained by using parallel processing techniques to test the resulting sub-images at the same time using the same number of fast neural networks. Moreover, the problem of sub-image centering and normalization in the Fourier space is solved.

Fast time delay neural networks.

This paper presents a new approach to speed up the operation of time delay neural networks. The entire data are collected together in a long vector and then tested as a one input pattern. The proposed fast time delay neural networks (FTDNNs) use cross correlation in the frequency domain between the tested data and the input weights of neural networks. It is proved mathematically and practically that the number of computation steps required for the presented time delay neural networks is less than that needed by conventional time delay neural networks (CTDNNs). Simulation results using MATLAB confirm the theoretical computations.

Fast Iris Detection for Personal Verification Using Modular Neural Nets

In this paper, a combination of fast and cooperative modular neural nets to enhance the performance of the detection process is introduced. I have applied such approach successfully to detect human faces in cluttered scenes [10]. Here, this technique is used to identify human irises automatically in a given image. In the detection phase, neural nets are used to test whether a window of 20×20 pixels contains an iris or not. The major difficulty in the learning process comes from the large database required for iris / noniris images. A simple design for cooperative modular neural nets is presented to solve this problem by dividing these data into three groups. Such division results in reduction of computational complexity and thus decreasing the time and memory needed during the test of an image. Simulation results for the proposed algorithm show a good performance. Furthermore, faster iris detection is obtained through image decomposition into many sub-images and applying cross correlation in frequency domain between each sub-image and the weights of the hidden layer.

New fast normalized neural networks for pattern detection

Neural networks have shown good results for detecting a certain pattern in a given image. In this paper, fast neural networks for pattern detection are presented. Such processors are designed based on cross correlation in the frequency domain between the input image and the input weights of neural networks. This approach is developed to reduce the computation steps required by these fast neural networks for the searching process. The principle of divide and conquer strategy is applied through image decomposition. Each image is divided into small in size sub-images and then each one is tested separately by using a single fast neural processor. Furthermore, faster pattern detection is obtained by using parallel processing techniques to test the resulting sub-images at the same time using the same number of fast neural networks. In contrast to fast neural networks, the speed up ratio is increased with the size of the input image when using fast neural networks and image decomposition. Moreover, the problem of local sub-image normalization in the frequency domain is solved. The effect of image normalization on the speed up ratio of pattern detection is discussed. Simulation results show that local sub-image normalization through weight normalization is faster than sub-image normalization in the spatial domain. The overall speed up ratio of the detection process is increased as the normalization of weights is done offline.

A New Implementation for High Speed Normalized Neural Networks in Frequency Space

Neural networks have shown good results for detection of a certain pattern in a given image. In our previous work, a fast algorithm for object/face detection was presented. Such algorithm was designed based on cross correlation in the frequency domain between the input image and the weights of neural networks. In this paper, a simple design for solving the problem of local subimage normalization in the frequency domain is presented. This is done by normalizing the weights in the spatial domain off line. Furthermore, it is proved that local subimage normalization by normalizing the weights is faster than subimage normalization in the spatial domain. Moreover, the overall speed up ratio of the detection process is increased as the normalization of weights is done off line.