Catálogo de publicaciones - libros

An efficient rotation invariant feature extraction technique for texture classification based on Gabor multi-channel filtering is proposed. In this technique, Gabor function is approximated by a set of steerable basis functions, which results in a significant saving in the computation cost. The classification of 15 classes of Brodatz textures are considered in our experiments. Results show that up to 40% of computation can be saved compared with traditional Gabor multi-channel filtering method. In the mean time, almost the same high texture classification correct rate can be achieved.

Multiresolution histograms have been recently proposed as robust and efficient features for texture classification. In this paper, we evaluate the performance of multiresolution histograms for texture classification using support vector machines (SVMs). We observe that the dimensionality of multiresolution histograms can be greatly reduced with a Laplacian pyramidal decomposition. With an appropriate kernel, we show that SVMs significantly improve the performance of multiresolution histograms compared to the previously used nearest-neighbor (NN) classifiers on a texture classification problem involving Brodatz textures. Experimental results indicate that multiresolution histograms in conjunction with SVMs are also robust to noise.

Texture classification plays an important role in identifying objects. The fractal properties based on moment feature images for texture classification are investigated in this paper. The two-order moments of the image in small windows are used as feature images whose fractal dimensions are then computed and employed to classify the textures using support vector machines (SVMs). Experiments on several Brodatz nature images and four in-vivo B-mode ultrasound liver images demonstrate the effectiveness of the proposed algorithm.

The paper presents a 2 frame structure-from-motion algorithm that operates by mapping local changes (image deformations) into estimates of time-to-collision (TTC). For constant velocity motion of the camera in a stationary scene, time-to-collision amounts to coarse depth data – useful for navigation and qualitative scene understanding. The theory is supported by a set of experiments demonstrating accurate TTC recovery from video sequence data acquired by a mobile robot.

Traditional motion segmentation techniques generally depend on a pre-estimated optical flow. Unfortunately, the lack of precision over edges of most popular motion estimation methods makes them unsuited to recover the exact shape of moving objects. In this contribution, we present an original motion segmentation technique using a -nearest-neighbor-based fusion of spatial and temporal label cues. Our fusion model takes as input a spatial segmentation of a still image and an estimated version of the motion label field. It minimizes an energy function made of spatial and temporal label cues extracted from the two input fields. The algorithm proposed is intuitive, simple to implement and remains sufficiently general to be applied to other segmentation problems. Furthermore, the method doesn’t depend on the estimation of any threshold or any weighting function between the spatial and temporal energy terms, as is sometimes required by energy-based segmentation models. Experiments on synthetic and real image sequences indicate that the proposed method is robust and accurate.

In mineral processing industry, it is often useful to be able to obtain statistical information about the size distribution of ore fragments that move relatively to a static but noisy background. In this paper, we introduce a novel approach to estimate the 2D shapes of multiple moving objects in noisy background. Our approach combines adaptive Gaussian mixture model (GMM) for background subtraction and optical flow methods supported by temporal differencing in order to achieve robust and accurate extraction of the shapes of moving objects. The algorithm works well for image sequences having many moving objects with different sizes as demonstrated by experimental results on real image sequences.

In this paper, we present a novel approach for analyzing the dynamic water motion and transforming the motion in natural scenes to non-photorealistic 2D cartoons. We limit the domain of the original water sequences containing only water surfaces with shallow relief, (the heights of the water wave are relatively small compared to the wave lengths) and one parallel light source. Within this constrained domain, we first automatically rectify the water wave sequence from a generic pose to orthogonal view direction. Then we clearly reveal the relationship between the rectified frames and the surface normal maps. Finally, as an application, a non-photorealistic rendering step is applied to transform the water motion to new cartoon sequences. several results are shown in the paper to demonstrate the quality and widely usability of this novel approach.

Skin detection has been employed in various applications including face and hand tracking, and retrieving people in video databases. However most of the currently available algorithms are either based on static features of the skin color, or require a significant amount of computation. Moreover, skin detection algorithms are not robust enough to deal with real-world conditions, such as background noise, change of intensity and lighting effects. This situation can be improved by using dynamic features of the skin color in a sequence of images. This article proposes a skin detection algorithm based on in-motion pixels of the image. The membership measurement function for recognizing skin/non skin is based on the Hue histogram of skin pixels that adapts itself to the user’s skin color, in each frame. This algorithm has demonstrated significant improvement in comparison to the static skin detection algorithms.

An efficient algorithm to segment the moving object is very important in the surveillance system. In general, the change detection by comparing brightness value is a good and simple method, but it shows a poor performance under illumination change. Therefore, we propose the segmentation algorithm to extract effectively the object in spite of the illumination change. There are three modes to extract the object, the criteria of mode selection are both available background existence and illumination change. Then the object is finally obtained by using projection and the morphological operator in post-processing. Furthermore, the double binary method using the similarity of brightness value and spatial proximity is used to obtain more edge information. A good segmentation performance is demonstrated by the simulation result.

This paper presents a robust approach to track multiple objects for low resolution, far-field visual surveillance applications. Multiple moving objects are detected by utilizing an adaptive background model and tracked by resolving the correspondence between their trajectory segments using proximity and appearance similarity measures. A new confidence measure is assigned to each possible match between objects and this information is maintained by a graph structure. This graph is utilized to prune and refine the trajectories. Kalman filter is used to handle discontinuities and occlusions. Proposed approach handles problems such as spurious objects, fragmentation, shadow, clutter and occlusions.