Hala Mousher Ebied

Evaluation of unsupervised feature extraction neural networks for face recognition

In this paper, new appearances based on neural networks (NN) algorithms are presented for face recognition. Face recognition is subdivided into two main stages: feature extraction and classifier. The suggested NN algorithms are the unsupervised Sanger principal component neural network (Sanger PCNN) and the self-organizing feature map (SOFM), which will be applied for features extraction of the frontal view of a face image. It is of interest to compare the unsupervised network with the traditional Eigenfaces technique. This paper presents an experimental comparison of the statistical Eigenfaces method for feature extraction and the unsupervised neural networks in order to evaluate the classification accuracies as comparison criteria. The classifier is done by the multilayer perceptron (MLP) neural network. Overcoming of the problem of the finite number of training samples per person is discussed. Experimental results are implemented on the Olivetti Research Laboratory database that contains variability in expression, pose, and facial details. The results show that the proposed method SOFM/MLP neural network is more efficient and robust than the Sanger PCNN/MLP and the Eigenfaces/MLP, when used a few number of training samples per person. As a result, it would be more applicable to utilize the SOFM/MLP NN in order to accomplish a higher level of accuracy within a recognition system.

Color Image Segmentation Based on Different Color Space Models Using Automatic GrabCut

This paper presents a comparative study using different color spaces to evaluate the performance of color image segmentation using the automatic GrabCut technique. GrabCut is considered as one of the semiautomatic image segmentation techniques, since it requires user interaction for the initialization of the segmentation process. The automation of the GrabCut technique is proposed as a modification of the original semiautomatic one in order to eliminate the user interaction. The automatic GrabCut utilizes the unsupervised Orchard and Bouman clustering technique for the initialization phase. Comparisons with the original GrabCut show the efficiency of the proposed automatic technique in terms of segmentation, quality, and accuracy. As no explicit color space is recommended for every segmentation problem, automatic GrabCut is applied with RGB, HSV, CMY, XYZ, and YUV color spaces. The comparative study and experimental results using different color images show that RGB color space is the best color space representation for the set of the images used.

Multi-label automatic GrabCut for image segmentation

This paper presents a multi-label automatic GrabCut technique for the problem of image segmentation. GrabCut is considered as one of the binary-label segmentation techniques because it is based on the famous s/t graph cut minimization technique for image segmentation. This paper extends the automatic binary-label GrabCut to a multi-label technique that can segment a given image into its natural segments without user intervention. Since multi-label segmentation is an NP-hard problem, the proposed algorithm converts the segmentation problem into multiple iterative piecewise binary label GrabCut segmentations. This implies separating one segment from the image, under consideration, per iteration. In this way, the proposed algorithm maintains the powerful advantage of the GrabCut to get the optimal solution for the segmentation problem. Evaluation of the segmentation results was carried out using different accuracy metrics from the literature. The evaluations were conducted with human ground truth segmentations from Berkeley benchmark dataset of natural images. Although human segmentations are semantically more meaningful, experiments showed that the proposed multi-label GrabCut provided matching segmentation results to that of individual humans with acceptable accuracy.

A Comparative Study of Different Color Space Models Using FCM-Based Automatic GrabCut for Image Segmentation

GrabCut is one of the powerful color image segmentation techniques. One main disadvantage of GrabCut is the need for initial user interaction to initialize the segmentation process which classifies it as a semi-automatic technique. The paper presents the use of Fuzzy C-means clustering as a replacement of the user interaction for the GrabCut automation. Several researchers concluded that no single color space model can produce the best results of every image segmentation problem. This paper presents a comparative study of different color space models using automatic GrabCut for the problem of color image segmentation. The comparative study includes the test of five color space models; RGB, HSV, XYZ, YUV and CMY. A dataset of different 30 images are used for evaluation. Experimental results show that the YUV color space is the one generating the best segmentation accuracy for the used dataset of images.

Hybrid cluster of multicore CPUs and GPUs for accelerating hyperspectral image hierarchical segmentation

Hierarchical image segmentation is a well-known image analysis and clustering method that is used for hyperspectral image analysis. This paper introduces a parallel implementation of hybrid CPU/GPU for the Recursive Hierarchical Segmentation method (RHSEG) algorithm, in which CPU and GPU work cooperatively and seamlessly, combining benefits of both platforms. RHSEG is a method developed by National Aeronautics and Space Administration (NASA) which is more efficient than other traditional methods for high spatial resolution images. The RHSEG algorithm is also implemented on both GPU cluster and hybrid CPU/GPU cluster and the results are compared with the hybrid CPU/GPU implementation. For single hybrid computational node of 8 cores, a speedup of 6x is achieved using both CPU and GPU. On a computer cluster of 16 hybrid CPU/GPU nodes, an average speed up of 112x times is achieved over the sequential CPU implementation.

Kernel-PCA for face recognition in different color spaces

Different color spaces are better for different applications. This paper investigates the performance of face recognition with some color spaces using kernel-based Principal Component Analysis (Kernel-PCA). Kernel-PCA is a non-linear extension from the popular algorithm PCA. Experiments are performed with the Gaussian kernel function. Color spaces are linear or non-linear transform from RGB. In this paper, the RGB, YCbCr, and HSV color spaces are compared with the gray image (luminance information Y). Kernel-PCA is used to extract features from individual color components or from combining the three components of every color space in one vector. The experiments are performed on FEI color database. FEI database is frontal face images with seven profile images rotation of up to about 180 degrees and two different facial expression images. The experimental results show that the V color component of the HSV color space outperform all the used color organization.

Automatic GrabCut for Bi-label Image Segmentation Using SOFM

This paper proposes a new technique for the problem of color image segmentation using GrabCut. GrabCut is considered as one of the semi-automatic segmentation techniques, since it requires user interaction for the initialization of the segmentation process, via dragging a rectangle around an object to extract it. This restricts GrabCut for bi-label segmentation, where the image cannot be segmented into more than two; foreground and background segments. In order to set up for multi-label segmentation, this paper presents the use of SOFM as a powerful unsupervised clustering technique for the GrabCut initialization process. This converts the GrabCut from a semi-automatic into a complete automatic segmentation technique. The use of different SOFM architectures for the process of image segmentation was tested for real experiments. Evaluation and comparison with the original GrabCut show the efficiency of the proposed automatic technique in terms of segmentation quality and accuracy.

Human action recognition via multi-scale 3D stationary wavelet analysis

Multi-scale methods, especially wavelets, are being used in various computer vision applications, including surveillance, robotics, and human-centered computing. Human action recognition is one of the core areas that dominate the aforementioned applications. In this paper, the 3D multi-scale stationary wavelet analysis is used to build a view-based multi-scale spatio-temporal representation of the human actions. The proposed representation benefits from the ability of the 3D stationary wavelet transform to fuse the spatio-temporal information highlighted at different scales and orientations. Experimental results using Weizmann and KTH datasets revealed a good performance in various scenarios with different conditions.

Satellite Super Resolution Image Reconstruction Based on Parallel Support Vector Regression

Super Resolution (SR) refers to the reconstruction of a high resolution image from one or more low resolution images for the same scene. The reconstruction process is considered an inverse problem to the observation model. In this paper the SR problem is formulated by using Support Vector Regression (SVR). SVR is a very expensive computationally algorithm, thus it could be accelerated by using the computational power of a Graphics Processing Unit (GPU). The proposed parallel SVR has been implemented using NVidia’s compute device unified architecture (CUDA). An experiment has been done for a real satellite image. The experimental result demonstrates the speedup of the presented GPU implementation and compared with the serial CPU implementation and state-of-the-art techniques. The speedup of the presented SVR GPU-based implementation is up to approximately 50 times faster than the corresponding optimized CPU.

Parallel Super-Resolution Reconstruction Based on Neighbor Embedding Technique

Super Resolution (SR) is a technique to recover a high-resolution (HR) image from different noisy low resolution (LR) images. The missing high-frequency components in LR images should be restored correctly in HR image. Because of the extensive size of satellite images, the utilize to parallel algorithms can accomplish results more quickly with accurate results. This paper proposes an accelerated parallel implementation for an example based super-resolution algorithm, Neighbor Embedding (NE), using GPU. The NE trains the dictionary with patches obtained from a single image in the training phase. Euclidean distances are used to obtain the optimal weights that will be used in the construction of high-resolution images. Compute Device Unified Architecture (CUDA) by NVidia’s has been used to implement the proposed parallel NE. Different experiments have been carried out on a synthetic test image and satellite test image. The proposed GPU implementation of the NE was benchmarked against the serial implementation. The experimental results show that the speed of the implementation depends on the image size. The speed of the GPU implementation compared to the serial one using CPU ranged from 20× for small images to more than 30× for large image size.