Catálogo de publicaciones - libros

In this paper, we propose an ownership identification scheme for digital images with binary and gray-level ownership statements. The proposed method uses the theories and properties of sampling distribution of means to satisfy the requirements of robustness and security. Essentially, our method will not really insert the ownership statement into the host image. Instead, the ownership share will be generated by the sampling method as a key to reveal the ownership statement. Besides, our method allows ownership statements to be of any size and avoids the hidden ownership statement to be destroyed by the latter ones. When the rightful ownership of the image needs to be identified, our method can reveal the ownership statement without resorting to the original image. Finally, several common attacks to the image will be held to verify the robustness and the security is also analyzed.

Accurate rigid and non-rigid tracking of faces is a challenging task in computer vision. Recently, appearance-based 3D face tracking methods have been proposed. These methods can successfully tackle the image variability and drift problems. However, they may fail to provide accurate out-of-plane face motions since they are not very sensitive to out-of-plane motion variations. In this paper, we present a framework for fast and accurate 3D face and facial action tracking. Our proposed framework retains the strengths of both appearance and 3D data-based trackers. We combine an adaptive appearance model with an online stereo-based 3D model. We provide experiments and performance evaluation which show the feasibility and usefulness of the proposed approach.

For the robust face detection, illumination is considered as one of the great challenges. Motivated with the adaptation of the human vision system, we propose the curve mapping (CM) function to adjust the illumination conditions of the images. The lighting parameter of CM function is determined by the intensity distribution of the images. Therefore the CM function can adjust the images according to their own illumination conditions adaptively. The CM method will abandon no information of the original images and bring no noises to the images. But it will enhance the details of the images and adjust the images to the proper brightness. Consequently the CM method will make the images more discriminative. Experimental results show that it can improve the performance of the face detection with the CM method as a lighting-filter.

In this paper, we present a new scheme called Common Image method for face recognition. Our method has a couple of advantages over the conventional face recognition algorithms; one is that it can deal with the Small Sample Size(SSS) problem in LDA, and the other one is that it can achieve a better performance than traditional PCA by seeking the optimal projection vectors from image covariance matrix in a recognition task. As opposed to traditional PCA-based methods and LDA-based methods which employ Euclidean distance, Common Image methods adopted Assemble Matrix Distance(AMD) and IMage Euclidean Distance(IMED), by which the overall recognition rate could be improved. To test the recognition performance, a series of experiments were performed on CMU PIE, YaleB, and FERET face databases. The test results with these databases show that our Common Image method performs better than Discriminative Common Vector and 2DPCA-based methods.

In this paper we propose the use of Discrete Choice Analysis (DCA) for static facial expression classification. Facial expressions are described with expression descriptive units (EDU), consisting in a set of high level features derived from an active appearance model (AAM). The discrete choice model (DCM) is built considering the 6 universal facial expressions plus the neutral one as the set of the available alternatives. Each alternative is described by an utility function, defined as the sum of a linear combination of EDUs and a random term capturing the uncertainty. The utilities provide a measure of likelihood for a combinations of EDUs to represent a certain facial expression. They represent a natural way for the modeler to formalize her prior knowledge on the process. The model parameters are learned through maximum likelihood estimation and classification is performed assigning each test sample to the alternative showing the maximum utility. We compare the performance of the DCM classifier against Linear Discriminant Analysis (LDA), Generalized Discriminant Analysis (GDA), Relevant Component Analysis (RCA) and Support Vector Machine (SVM). Quantitative preliminary results are reported, showing good and encouraging performance of the DCM approach both in terms of recognition rate and discriminatory power.

This paper considers the design of an optimal joint source-channel coding system employing scalable wavelet-based source coders and unequal error protection for error-resilient transmission over binary erasure channels. We theoretically show that the expected average rate-distortion function in the separately encoded subbands is convex with monotonically decreasing slopes and use this fact to propose a new algorithm which minimizes the end-to-end distortion. This algorithm is based on a simplified Viterbi-search followed by Lagrangian-optimization. We show that our proposed solution results in significant complexity reductions while providing very near-to-optimal performance.

We address the topic of real-time analysis and recognition of silhouettes. The method that we propose first produces object features obtained by a new type of morphological operators, which can be seen as an extension of existing granulometric filters, and then insert them into a tailored classification scheme.

Intuitively, given a binary segmented image, our operator produces the set of all the largest rectangles that can be wedged inside any connected component of the image. The latter are obtained by a standard background subtraction technique and morphological filtering. To classify connected components into one of the known object categories, the rectangles of a connected component are submitted to a machine learning algorithm called EXtremely RAndomized trees (Extra-trees). The machine learning algorithm is fed with a static database of silhouettes that contains both positive and negative instances. The whole process, including image processing and rectangle classification, is carried out in real-time.

Finally we evaluate our approach on one of today’s hot topics: the detection of human silhouettes. We discuss experimental results and show that our method is stable and computationally effective. Therefore, we assess that algorithms like ours introduce new ways for the detection of humans in video sequences.

This paper proposes a new approach in digital watermarking applications that can be adapted for embedding either fragile or robust watermarking in a digital image in the spatial domain or in the frequency domain. The main objective of the proposed scheme is to explore the amount of texture or edge pixels belonging to the host image in order to insert more information while preserving the robustness of the scheme without degrading the visual quality of the watermarked image. The host image is divided into blocks and each block can be subdivided into sub-blocks according to its texture analysis. The number of sub-blocks that each block will be divided into depends on the amount of texture or edge pixels presented by it. The numerical results show that the proposed scheme is better in JPEG compression attacks, and far exceeds others in watermark size capacity.

In this paper we propose a method that analyzes attack patterns and extracts watermark after restoring the watermarked image from the geometric attacks. The proposed algorithm consists of a spatial-domain key insertion part for attack analysis and a frequency-domain watermark insertion part using discrete wavelet transform. With the spatial-domain key extracted from the damaged image, the proposed algorithm analyzes distortion and finds the attack pattern. After restoring the damaged image, the algorithm extracts the embedded watermark. By using both spatial domain key and frequency domain watermark, the proposed algorithm can achieve robust watermark extraction against geometrical attacks and image compressions such as JPEG.

Skin detection consists in detecting human skin pixels from an image. In this paper we propose a new skin detection algorithm based on approximation of an image patch joint distribution, called Best-Tree distribution. A tree distribution model is more general than a bayesian network one. It can represent a joint distribution in an intuitive and efficient way. We assess the performance of our method on the Compaq database by measuring the Receiver Operating Characteristic curve and its under area. These measures have proved better performances of our model than the baseline one.