Face recognition is an interesting and a challenging problem that has been widely studied in the field of pattern recognition and computer vision. It has many applications such as biometric authentication, video surveillance, and others. In the past decade, several methods for face recognition were proposed. However, these methods suffer from pose and illumination variations. In order to address these problems, this paper proposes a novel methodology to recognize the face images. Since image gradients are invariant to illumination and pose variations, the proposed approach uses gradient orientation to handle these effects. The Schur decomposition is used for matrix decomposition and then Schurvalues and Schurvectors are extracted for subspace projection. We call this subspace projection of face features as Schurfaces, which is numerically stable and have the ability of handling defective matrices. The Hausdorff distance is used with the nearest neighbor classifier to measure the similarity between different faces. Experiments are conducted with Yale face database and ORL face database. The results show that the proposed approach is highly discriminant and achieves a promising accuracy for face recognition than the state-of-the-art approaches.

No abstract

In the past, several automatic video summarization systems had been proposed to generate video summary. However, a generic video summary that is generated based only on audio, visual and textual saliencies will not satisfy every user. This paper proposes a novel system for generating semantically meaningful personalized video summaries, which are tailored to the individual user's preferences over video semantics. Each video shot is represented using a semantic multinomial which is a vector of posterior semantic concept probabilities. The proposed system stitches video summary based on summary time span and top-ranked shots that are semantically relevant to the user's preferences. The proposed summarization system is evaluated using both quantitative and subjective evaluation metrics. The experimental results on the performance of the proposed video summarization system are encouraging.

Face recognition in deity images is a challenging problem. Most of the existing face recognition methods are very sensitive to pose and illumination changes. This paper proposes a new technique for deity face recognition which is suitable for pose and illumination changes. The proposed approach uses Schur decomposition to speedup PCA computations and doubly modified Hausdorff distance for measuring similarity between different face edge maps. In addition, this paper introduces a new dataset named as Indian DEity dataSet (IDES) for face recognition which contains a collection of face images of Indian deities. Performances of the proposed method for deity face recognition are experimented with IDES dataset. The results show that the proposed method provides promising face recognition accuracy with pose and illumination changes.

No abstract

This paper presents a new methodology for image description to extract discriminative features from images, that are robust and invariant to illumination changes. The proposed approach describes the image features using multisupport interest regions. The features from multisupport regions are extracted after transforming each interest region into log polar form. Then the Weber's law is utilized to enhance the log polar image features to improve discriminability power of the descriptor. To prove the efficiency of the proposed image descriptor, experiments are conducted using Oxford dataset and compared our descriptor with the standard image descriptors such as SIFT, GLOH and DAISY. The experimental results demonstrate the robustness of the proposed descriptor against image variations in illumination.

Local image region description is a fundamental task for image feature matching in the field of Computer Vision. A good image region descriptor should have the ability to discriminate image features even though the images differ due to photometric variations and geometric transformations. Over these years, many local region descriptors have been proposed to tackle the aforementioned challenges. Achieving rotation invariance in keypoint description is considered one of the main challenges in local region description and matching. Previous approaches proposed to tackle rotation variations depend on unreliable dominant orientation estimation. In this paper, two novel local image region descriptors called Local Intensity Order-based Center Symmetric Local Binary Patterns (LIOCSLBP) and Local Intensity Order-based Orthogonally Combined Local Binary Patterns (LIOOCLBP) are proposed to build rotation invariant local region descriptions. The rotation invariance characteristic of the proposed binary pattern-based local region description is achieved by applying a simple and efficient mechanism called Local Intensity Ordering (LIO). The proposed descriptors use double interest regions for each interest point to improve feature discrimination. In order to further improve the feature discrimination ability RGBLIOCSLBP, RGBLIOOCLBP, HSVLIOCSLBP and HSVLIOOCLBP are also proposed exploiting RGB and HSV color models. Extensive experiments are conducted to evaluate the performance of the proposed descriptors on standard benchmark datasets for image matching, object recognition and scene recognition against the state-of-the-art descriptors. The experimental results show that the proposed descriptors are highly competitive to several stateof-the-art local region descriptors where the proposed descriptors outperformed the comparative approaches in many cases.

No abstract