J. M. H. du Buf

Texture feature performance for image segmentation

Abstract This paper describes a comparative study of texture features, with particular emphasis on the applicability to unsupervised image segmentation. A benchmark test is introduced in which a set of 20 simple bipartite images, combining different stochastic textures separated by a stochastic boundary, is used for feature extraction and segmentation. The accuracy of the segmentation result, expressed in the mean boundary error, is used as an evaluation criterion. From the seven feature extraction methods tested, the Haralick, Laws and Unser methods gave best overall results. Results obtained also show that direct feature statistics such as the Bhattacharyya distance are not appropriate evaluation criteria if texture features are used for image segmentation. A small experiment on visual boundary tracking revealed that boundary error obtained here are similar to those obtained by machine segmentation.

N-folded symmetries by complex moments in Gabor space and their application to unsupervised texture segmentation

Keywords: LTS1 Reference LTS-ARTICLE-1994-002 Record created on 2006-06-14, modified on 2016-08-08

SIMULTANEOUS DETECTION OF LINES AND EDGES USING COMPOUND GABOR FILTERS

Lines and edges play an important role in pattern recognition because they mark object surfaces and boundaries. In spite of many attempts to construct optimal detectors, e.g. an edge detector, it appears that all known algorithms have problems at locations where lines and edges are very close and/or intersect. Furthermore, there are still very few schemes which can detect and classify lines and edges (events) simultaneously. In view of the fact that the human visual system does not seem to suffer from any problems, our aim is to develop an event detection scheme that makes use of biologically-motivated operators and, therefore, to overcome the problems known from the literature. In this scheme we apply 2D complex Gabor filters and exploit the local phase information. In order to cope with events having a different curvature, we develop an adaptive scheme that is based on compound Gabor filter kernels with different orientation bandwidths.

A quantitative comparison of edge-preserving smoothing techniques

Abstract Edge-preserving smoothing techniques are compared by considering a test image which contains a central disk-shaped region with a step or a ramp edge against a uniform background. Free parameters are the amplitude of Gaussian noise added, the edge slope and the number of filtering iterations. The quantitative comparison measure is the normalised squared error between the filtered noisy image and the noise-free image, on the uniform image regions and on the transition region separately. The filters considered are analysed with respect to their performance under variations in the free parameters and their computer-time consumption. Results obtained are compared with published data available.

Gabor phase in texture discrimination

Abstract The relevance of the local phase information for texture discrimination and image segmentation is studied. Useful local phase information is obtained by unwrapping the phase images which result from the application of 2D Gabor filters, i.e., with both a frequency and an orientation selectivity. The unwrapping method applied is based on phase demodulation, where the global linear phase component in the orientation of the filter is directly measured from the wrapped phase image. Experimental results seem to confirm the importance of the local phase information.

Texture features based on Gabor phase

Abstract Local phase information obtained by Gabor filtering may be used for texture discrimination and image segmentation purposes. Explored are the densities of isolated zero points, discontinuities and zero contours. The phase gradient and local demodulation methods are studied as well. It is shown that most of these methods can render useful boundary and region information if simple test images are considered. However, the quality of this information degrades if the textures are disturbed by a small amount of jitter or by band-limited additive noise.

Abstract processes in texture discrimination.

In this study some experiments on texture segmentation are reported using the local Gabor power spectrum. The techniques applied are: (1) supervised pixel classification; (2) boundary detection by spectral dissimilarity estimation; (3) region-based segmentation based on Gaussian spectral estimation; and (4) the same as (3) but based on central moments of the local spectrum. It is shown that very-acceptable-to-excellent results can be obtained. It is argued, however, that the shortcomings of region-based and boundary-based approaches require that both processes should act in parallel, not only in digital image processing but also in the modelling of visual perception.

Realtime local navigation for the blind: detection of lateral doors and sound interface

Worldwide there are about 285 million visually impaired persons, of which 39 million are blind and the others have low vision. Almost all systems designed to assist them are quite complex and expensive, but most blind persons do not have advanced technical assistance and they are rather poor. We are therefore developing a low-cost navigation aid which can be afforded by almost all blind persons: basically, the ultimate goal is to use only a mobile phone with a built-in camera. This aid complements the white cane, it is easily portable, and it is not a hindrance when walking with the cane. The system will have an easy and intuitive interface, yet providing assistance in local and global navigation in realtime. In this paper we present the progress concerning local navigation. Path and obstacle detection just beyond the reach of the cane is now supplemented by detection of doors in corridors. This is necessary for localization, i.e., for developing a better impression of the environment and for finding a specific room. A sophisticated sound interface can assist the user for centering on paths like sidewalks and corridors, alerting to looming obstacles for avoiding them.

Indoor Localization and Navigation for Blind Persons using Visual Landmarks and a GIS

In an unfamiliar environment we spot and explore all available information which might guide us to a desired location. This largely unconscious processing is done by our trained sensory and cognitive systems. These recognize and memorize sets of landmarks which allow us to create a mental map of the environment, and this map enables us to navigate by exploiting very few but the most important landmarks stored in our memory. We present a system which integrates a geographic information system of a building with visual landmarks for localizing the user in the building and for tracing and validating a route for the users navigation. Hence, the developed system complements the white cane for improving the users autonomy during indoor navigation. Although designed for visually impaired persons, the system can be used by any person for wayfinding in a complex building.

Improved grating and bar cell models in cortical area V1 and texture coding

This paper presents improved models of cortical neurons in V1 that act like grating and bar detectors. Both models use the same frontend, which consists of a contrast normalisation in combination with isotropic DOG filtering, followed by anisotropic Gabor filtering together with a response sharpening. Different grouping processes of ON and OFF responses lead to a very selective detection of bar width and grating frequency, with a good localisation. Furthermore, outputs of grating cells can be grouped over combinations of orientations for coding nonlinear textures. It is shown that these models, apart from being used in the modelling of the visual cortex, can be employed in pattern-recognition applications.