Catálogo de publicaciones - libros

In this paper we present our region merging algorithm that is built with special attention on speed but still includes all the necessary functionality to use a wide range of both region based and border based dissimilarity criteria. The algorithm includes a novel method to dynamically link the common borders between two segments during the region merging. The main part of the paper will concentrate on the efficient data structures and functions that are needed to obtain a fast region merging algorithm. Also, all the special situations that can occur in the segment topology are completely covered. We give a detailed report on the execution times of the algorithm and show some of the segmentation results we obtained.

Due to the variable block sizes and multi-reference frame used in the H.264/AVC standard; the motion estimation process becomes even more computationally intensive, resulting in a very low encoding speed. To overcome this low encoding speed, fast motion estimation algorithms such as UMHexagonS [1] and EPZS [2] have been proposed. Since the integer-pixel motion estimation speed has significantly decreased, the fractional or sub-pixel motion estimation speed is no longer non-negligible. We propose a fast sub-pixel motion estimation algorithm using an adaptive rood pattern based on the fractional motion vector of adjacent blocks and also a simplified small diamond search. Our algorithm is able to reduce the number of sub-pixel search points by more than 50%, while restricting the PSNR loss to less than 0.1 dB, compared to the hierarchical fractional pixel search.

The H.264/AVC standard uses new coding tools to improve coding efficiency. Among the tools, motion estimation using smaller block sizes leads to higher correlation between the motion vectors of neighboring blocks. This characteristic of H.264/AVC is useful for motion vector recovery to conceal a lost macroblock. In this paper, we propose a motion vector recovery method based on optical flow in H.264/AVC video coding. We first determine the optical flow region to alleviate the complexity, and choose an initial value of flow velocity using neighboring motion vectors of a lost macroblock. The proposed method recovers the motion vectors of 4x4 blocks included in a lost macroblock using the weighted average of obtained flow velocities. Simulation results show that our proposed method gives higher objective and subjective visual qualities than conventional approaches.

It is well known that motion detection using single frame differencing, while computationally much simpler than other techniques, is more liable to generate large areas of false foregrounds known as . In order to overcome this problem the authors propose a method based on signed differencing and connectivity analysis. The proposal is suitable to applications which cannot afford the un-avoidable errors of background modeling or the limitations of 3-frames preprocessing.

A moving object in a video sequence is removed and corresponding background is completed by using a novel global feature registration technique. To find a 2D homography between two adjacent video frames, we track background and foreground features, separately. After estimating the homography, we extract and remove the moving object in every frame. To fill the background of the removed object accurately, we introduce a global feature registration technique. The technique iteratively reduces and distributes the accumulation errors associated to global video registration. Experimental results show that the proposed technique yields seamless background sequences.

This paper presents an approach for evaluating multiple color histograms during object tracking. The method adaptively selects histograms that well distinguish foreground from background. The variance ratio is utilized to measure the separability of object and background and to extract top-ranked discriminative histograms. Experimental results demonstrate how this method adapts to changing appearances of both object undergoing tracking and surrounding background. The advantages and limitations of the particle filter with embedded mechanism of histogram selection are demonstrated in comparisons with the standard CamShift tracker and a combination of CamShift with histogram selection.

This paper presents an implementation of a color-based multiple agent tracking algorithm targeted for wireless image sensor networks. The proposed technique is based on employing lightweight algorithms and low-bandwidth data communication between multiple network nodes to track the path of autonomous agents moving across the fields of view (FOV) of the sensors. Segmentation techniques are applied to find the agents within the FOV, and a color histogram is constructed using the hue values of the pixels corresponding to agents. This histogram is used as a means of identification within the network. As such, the algorithm is able to reliably track multi-colored agents of irregular shapes and sizes and can resolve identities after collisions. The proposed algorithm has low computational requirements and its complexity scales linearly with the size of the network, so it is feasible in low-power, large-scale wireless sensor networks.

A fast motion estimation (ME) algorithm is proposed to search motion vectors for variable blocks. The proposed method is based on the successive elimination algorithm (SEA) using sum norms to find the best estimate of the motion vectors and to implement efficient calculations for variable blocks. The proposed ME algorithm is applied to the Joint Video Team (JVT) encoder that performs a variable-block ME. In terms of computational complexity, the proposed ME algorithm with limited search range searches motion vectors at about 6.3 times as fast as the spiral full search and 5.5 times as fast as the fast full search using the hierarchical sum of absolute difference (SAD), while the PSNR (peak signal-to-noise ratio) of the reconstructed image is slightly degraded with only 0.1~0.4 dB.

Most existing object segmentation algorithms suffer from a so-called under-segmentation problem, where parts of the segmented object are missing and holes often occur inside the object region. This problem becomes even more serious when the object pixels have similar intensity values as that of backgrounds. To resolve the problem, we propose a constrained region-growing and contrast enhancement to recover those missing parts and fill in the holes inside the segmented objects. Our proposed scheme consists of three elements: (i) a simple linear transform for contrast enhancement to enable stronger edge detection; (ii) an 8-connected linking regional filter for noise removal; and (iii) a constrained region-growing for elimination of those internal holes. Our experiments show that the proposed scheme is effective towards revolving the under-segmentation problem, in which a representative existing algorithm with edge-map based segmentation technique is used as our benchmark.

Motion estimation by means of spatio-temporal energy filters –velocity tuned filters– is known to be robust to noise and aliasing and to allow an easy treatment of the aperture problem. In this paper we propose a motion representation based on the composition of spatio-temporal energy features, i.e., responses of a set of filters in phase quadrature tuned to different scales and orientations. Complex motion patterns are identified by unsupervised cluster analysis of energy features. The integration criterion reflects the degree of alignment of maxima of the features’s amplitude, which is related to phase congruence. The composite-feature representation has been applied to motion segmentation with a geodesic active model both for initialization and image potential definition. We will show that the resulting method is able to handle typical problems, such as partial and total occlusions, large inter-frame displacements, moving background and noise.