Search

Burned forest regions have to be monitored for several years after the fire so that the regeneration processes can be evaluated. The authors assess the danger of desertification conditions ensuing in the site of a burned forest in the Mediterranean region. If the forest does not show signs of recovery a couple of years after the fire, it probably has to be artificially re-forested to prevent further erosion. Quite often, different types of vegetation grow in a burned region. Usually this new vegetation presents a deterioration of the quality of the flora of the region, e.g. if bushes develop instead of trees. The main type of forests that are common in the Mediterranean region consist of Aleppo pine (Pinus Halepensis). The authors assess the degree of presence of three classes in a region: Aleppo pine, bare soil and other vegetation, using Landsat TM images. The authors show how the Hough transform can be used to identify the composition of a given region. The method is first assessed using simulated data and then it is tested using real data that have been obtained in the laboratory.

In segmenting an image by pixel classification, the sequence of gray levels of the pixel's neighbors can be used as a feature vector. This yields classifications at least as good as those obtained using other local properties (e.g., averages or values of difference operators) as features.

Two experimental studies of estimating land use classification in LANDSAT imagery by use of iterative relaxation operators are presented. In one of the experiments the effects of varying the number of classes are examined. In the other, classification accuracy in various kinds of terrain areas is studied.

A new algorithm for segmentation of images containing textured regions is presented. The algorithm is named MITES, which is an acronym for M odel-driven, I terative, T exture S egmentation. MITES represents an alternative to the traditional pixel classification approach to texture image segmentation because it makes explicit use of the spatial coherence of uniformly textured regions.

The paper describes the result of an experimental study concerning the use of probabilistic relaxation for improving pixel classification rates. Two LACIE sites were used in the study and in both cases, relaxation resulted in a marked improvement in classification rates.

Background subtraction models based on mixture of Gaussians have been extensively used for detecting objects in motion in a wide variety of computer vision applications. However, background subtraction modeling is still an open problem particularly in video scenes with drastic illumination changes and dynamic backgrounds (complex backgrounds). The purpose of the present work is focused on increasing the robustness of background subtraction models to complex environments. For this, we proposed the following enhancements: a) redefine the model distribution parameters involved in the detection of moving objects (distribution weight, mean and variance), b) improve pixel classification (background/foreground) and variable update mechanism by a new time-space dependent learning-rate parameter, and c) replace the pixel-based modeling currently used in the literature by a new space-time region-based model that eliminates the noise effect caused by drastic changes in illumination. Our proposed scheme can be implemented on any state of the art background subtraction scheme based on mixture of Gaussians to improve its resilient to complex backgrounds. Experimental results show excellent noise removal and object motion detection properties under complex environments.

Local granulometric size distributions are generated by performing a granulometry on an image and keeping local pixel counts in a neighborhood of each pixel at the completion of each successive opening. Normalization of the resulting size distributions yields a probability density at each pixel. These densities contain texture information local to each pixel. Pixels can be classified according to the moments of the densities. Further refinement can be accomplished by employing several structuring-element sequences in order to generate a number of granulometries, each revealing different texture qualities. Classification is accomplished by comparing observed moments to those representing a database of textures. The collection of database moments are actually random variables dependent on random texture processes, and the method employed in the present paper involves the comparison of observed moments to the means of database-texture moments.

A skin detection technique suitable for on-chip integration into color CMOS imagers is proposed. The technique uses color information in the normalized rgb color space to discriminate between skin and non-skin color pixels. Its current-mode implementation achieves real-time skin detection. Furthermore, the pixel classification is performed on read-out, allowing high-speed processing without the need for any on-chip memory elements. Fully analog, the proposed VLSI architecture combines compactness, programmability and low power operation.

Worldwide, more than 180 million people suffer from diabetes mellitus. Approximately 50% of these patients will develop complications to their feet. Neuropathy, combined with poor blood supply and biomechanical changes results in a high risk for foot ulcers, which is a key problem in the diabetic foot; when these wounds become infected, this can ultimately result in lower extremity amputation, which has a serious effect on the quality of life of the patient, and causes a large economic burden on society. This was the motivation for a collaborate project (Vincent50) in which a photographic foot imaging device was developed. The system allows scanning of the foot soles on a daily basis which may lead to early recognition of foot problems. The goal of the present study is to determine whether pixel classification is a useful intermediate step towards automatically assessing the images of the foot soles for signs of diabetic foot disease. If successful, this approach will further relief health care professionals in assessing the foot and enable the placement of more devices in the future. The best agreement between automated recognition and expert diagnosis was achieved with a combination of RGB and derived features, proves that the RGB data is informative with respect to detection of ulcers. However, the automatic detection of pre-signs of ulcers and other anomalies needs more sophistication than pixel classification alone. Firstly, other physical features, such as hyperspectral data, infrared and/or textural features are expected to be more informative. Secondly, we expect to be able to boost the performance by using the context between neighboring pixels. Thirdly, an individualized and normalized classification process might help with the large variability in foot soles between individuals.

Learning filters to produce sparse image representations in terms of over complete dictionaries has emerged as a powerful way to create image features for many different purposes. Unfortunately, these filters are usually both numerous and non-separable, making their use computationally expensive. In this paper, we show that such filters can be computed as linear combinations of a smaller number of separable ones, thus greatly reducing the computational complexity at no cost in terms of performance. This makes filter learning approaches practical even for large images or 3D volumes, and we show that we significantly outperform state-of-the-art methods on the linear structure extraction task, in terms of both accuracy and speed. Moreover, our approach is general and can be used on generic filter banks to reduce the complexity of the convolutions.

In this paper we empirically analyze the importance of sparsifying representations for classification purposes. We focus on those obtained by convolving images with linear filters, which can be either hand designed or learned, and perform extensive experiments on two important Computer Vision problems, image categorization and pixel classification. To this end, we adopt a simple modular architecture that encompasses many recently proposed models. The key outcome of our investigations is that enforcing sparsity constraints on features extracted in a convolutional architecture does not improve classification performance, whereas it does so when redundancy is artificially introduced. This is very relevant for practical purposes, since it implies that the expensive run-time optimization required to sparsify the representation is not always justified, and therefore that computational costs can be drastically reduced.

While learned image features can achieve great accuracy on different Computer Vision problems, their use in real-world situations is still very limited as their extraction is typically time-consuming. We therefore propose a method to learn image features that can be extracted very efficiently using separable filters, by looking for low rank filters. We evaluate our approach on both the image categorization and the pixel classification tasks and show that we obtain similar accuracy as state-of-the-art methods, at a fraction of the computational cost.

As part of the development of a sweet-pepper harvesting robot, obstacles should be detected. Objectives were to classify sweet-pepper vegetation into five plant parts: stem, top of a leaf (TL), bottom of a leaf (BL), fruit and petiole (Pet); and to improve classification results by post-processing. A multi-spectral imaging set-up with artificial lighting was developed to acquire images of sweet-pepper plants. The background was segmented from the vegetation and vegetation was classified into five plant parts, through a sequence of four two-class classification problems. True-positive detection rate/scaled false-positive rate achieved, on a pixel basis, were 40.0/179% for stem, 78.7/59.2% for top of a leaf (TL), 68.5/54.8% for bottom of a leaf (BL), 54.5/17.2% for fruit and 49.5/176.0% for petiole (Pet), before post-processing. The opening operations applied were unable to remove false stem detections to an acceptable rate. Also, many false detections of TL (>10%), BL (14%) and Pet (>15%) remained after post-processing, but these false detections are not critical for the application because these three plant parts are soft obstacles. Furthermore, results indicate that TL and BL can be distinghuished. Green fruits were post-processed using a sequence of fill-up, opening and area-based segmentation. Several area-based thresholds were tested and the most effective threshold resulted in a true-positive detection rate, on a blob basis, of 56.7 % and a scaled false-positive detection rate of 6.7 % for green fruits (N=60). Such fruit detection rates are a reasonable starting point to detect obstacles for sweet-pepper harvesting. But, additional work is required to complement the obstacle map into a complete representation of the environment.