Previous investigations indicate that edge information, gray level histogram information, texture information, and shape information are all useful in detecting objects. Gray level cooccurrence matrices contain a form of each of these types of information. Hence applying measures defined on cooccurrence matrices to the object detection problem would seem to be an approach which should be investigated. This paper presents a formulation of such an approach. The decision logic employed assumes that measures computed from regions containing the object form a cluster defined by N (μ 0 , Σ 0 ) in measurement space, while measures computed from regions not containing the object lie some distance away from this cluster. To help assure that this is the case, a measurement selection algorithm was formulated. Studies are reported which show testing detection accuracies of better than 90% on a number of experiments.

Object detection based on gray level cooccurrence

This paper presents some recent experimental results in robotic visual servoing. A general-purpose computer in conjunction with special purpose video processing hardware, in particular a newly available hardware region-growing and moment-generation unit, has been used to visually close the robot position loop at video field rates, 60Hz.

/pdf/video-rate-visual-servoing-for-robots-2hnx2usggf.pdf

Video- Rate Visual Servoing for Robots

A Radon-based invariant image analysis method is introduced. The linearity, shift, rotation, and scaling properties of the Radon transform are utilized to achieve invariant features to translation, rotation, and scaling. The singular values of a matrix, constructed by row-stacking of projections, are used to construct the invariant feature vector. This feature vector will be used as input to a classifier, which is here, the back-propagation neural network followed by a maximum-output-selector. A performance function is introduced to evaluate the performance of the recognition system. This performance function can also be used to indicate how closely the pattern matches the decision template. The effectiveness of this method is illustrated by a simulation example and it is compared with the method of Zernike moments.

Invariant image analysis based on Radon transform and SVD

In this review paper, a qualitative comparison of the performance of nine different segmentation algorithms on a data base of infrared images of vehicles is described. The segmentation methods are categorised according to their mode of operation into three distinct generic classes of algorithm: namely ‘grey level threshold techniques’, ‘three dimensional histogram methods’ and ‘pixel classification techniques’. Each segmentation technique is guided to a subset of the image by a spoke filter detection algorithm which locates regions of the scene that most resemble blob shaped man-made objects. A short list of four segmentation algorithms is compiled, of which two methods from the ‘pixel classification’ class, a K-nearest neighbour (KNN) and a Bayesian algorithm, are selected. The final preference is for the Bayesian technique, the KNN method being less favoured owing to the higher computational burden.

Comparison of segmentation processes for object acquisition in infrared images

In many computer vision systems accurate identification of various objects appearing in a scene is required. In this paper we address the problem of object detection in analyzing high resolution multispectral aerial images. Development of a practical object detection approach should consider issues of speed, accuracy, robustness, and amount of supervision allowed. The approach is based upon extraction of information from images and their systematic analysis utilizing available prior knowledge of various physical attributes of the objects. The step-wise approach examines spectral, spatial, and topographic features in making the object vs background decision. Techniques for the analysis of the spectral, spatial, and topographic features tend to be of increasing levels of computational complexity. The computationally simpler spectral feature analysis is performed for the entire image to detect candidate object regions. Only these regions are considered in the spatial feature analysis step to further reduce the number of candidate regions which need to be analyzed in the topographic feature analysis step. Such step-wise analysis makes the entire object detection process efficient by incorporating the process of “focus of attention” to identify regions of interest thus eliminating a relatively large portion of image from further detailed examination at every stage. Results of the experiments performed using several high resolution multispectral images have demonstrated the basic feasibility of the approach. The images utilized in the experiments are acquired from geographically different locations, at different times, with different types of background, and are of different resolution. Successful object detection with high accuracy and low false alarm rates indicate the robustness of this approach.

Object detection by step-wise analysis of spectral, spatial, and topographic features

An approach is described for detecting and classifying tactical targets in FLIR imagery. The basic assumption used for segmenting objects from their background is that the objects to be detected differ from the background in grey level, edge, properties, or texture. Potential targets are selected from a large frame, by locating combinations of grey level, edge, value, and texture that occur infrequently over the entire frame. Once potential objects are obtained, they are segmented from their backgrounds using the identical process as above, except applied on a local level. The segmented objects are classified into three, types of vehicles or into false, alarms. The classification procedure uses features measured on projections made through the segmented objects. Results are shown for 32 test images.© (1979) COPYRIGHT SPIE--The International Society for Optical Engineering. Downloading of the abstract is permitted for personal use only.

Stephen M. Lutton

Papers

Segmentation And Classification Of Targets In Flir Imagery

Adaptive Segmentation of Unique Objects