Spatial Entropy-based Cost Function for Gray and Color Image Segmentation with Dynamic Optimal Partitioning
MK Quweider · 2012
Summary In this paper, we present a novel thresholding-based segmentation algorithm that combines entropy, image spatial information, and dynamic programming to non-uniformly quantize an image in a more efficient and effective way for subsequent processing. Combined with information related to the structural content present in the image (activity/busyness of pixels with respect to their immediate neighbors), an entropy- based cost function is derived and used with the one-dimensional histogram probability distribution function of the image. The image quantization/ segmentation algorithm uses dynamic programming based on a recently introduced algorithm for optimal partitioning on an interval, and allow the selection of a broad range gray level to be present in the output image; binarization of an image is accomplished by having only two gray levels in the output image. Applications of the algorithm to quantization of gray-level as well as color images in the RGB and HSV color spaces are presented. Image simulations give very good results compared to many existing methods, while maintaining low computational complexity in terms of storage and processing requirements. industry. The HSV (hues, saturation, and value) color space model is used in applications based on perceptual properties of the human visual system. It is rare for an image processing system to work directly on the acquired image without first representing it in a more compact, economical, and efficient form (from a processing point of view), while keeping the structural and informational content intact. The step is done mainly for efficiency purposes. One way of achieving this is to reduce the number of gray levels present in the image (for gray-level images) or one or more of its components (for color images). Reducing the number of gray levels in an image is a fundamental issue in many image processing and computer vision applications including segmentation, thresholding, lossy compression, and image retrieval, just to name few. Segmentation is closely related to the gray level reduction problem. Although there exists many methods for segmentation, thresholding remains one of the most attractive and simple ones. Quantization can be seen as a multi-level thresholding problem, which is the view we adopt in this paper; when one-threshold is generated (two quantization levels) for the output image, the problem reduces to that of binarization, and when multiple thresholds are generated, the output image quantized gray levels can be judiciously selected between the thresholds to produce a pleasing and ready-to-process image. This paper presents a new multi-level thresholding method for image segmentation that allows the user to reduce the number of gray levels, in a hierarchical fashion, from the original number present in the image all the way down to two gray levels, corresponding to a binary version of the image. Our contribution to the problem is twofold: first, we present an entropy-based dynamic cost function that automatically adapts to the size of the region under examination so far; second, we integrate the cost function seamlessly with an interval optimal partitioning algorithm that uses dynamic programming. In section 2 and 3, we review thresholding an entropy-based thresholding respectively. In section 4 we briefly describe optimal partitioning; section 5 presents the entropy-based segmentation algorithm in details. Section 6 presents simulations with comparative results of some existing methods. Conclusions and future work are given in section 7.