Contrast enhancement technique based on local detection of edges

Many digital contrast enhancement techniques have been used in order to optimize the visual quality of the image for human or machine vision through gray-scale or histogram modifications [l-4]. All these methods try to enhance the contrast of the input image without measuring the contrast itself. Moreover there is no universal criterion or unifying theory specifying the validity of a given enhancement technique [2]. Contrast measures the relative decrease of the luminance in an image. It is highly correlated to the intensity gradient. Furthermore, the contrast can be deftned locally or globally. When an image is composed of textured regions it seems more adequate to perform a local analysis and thus to define a local contrast. In the present paper we only consider the local contrast associated to each pixel in a given neighborhood. It is well known that local contrast can be enhanced by computing the differences between the signal in each picture element and those in surrounding pixels and by amplifying these differences [5]. The original idea of Gordon et al. is based on these considerations. They directly define a contrast function and contrast enhancement functions without taking into account either the gray-scale or the gray level histogram of the input image. This is the reason why their method is more efficient and powerful than the classical ones. However, they tested their method only on mammographies [6,7]. The main deficiency of this method lies in the noise and the size of the details to be shown. The principal concern of this communication is to propose another definition of the contrast, associated to a pixel in a given neighborhood, according to some physical criteria in order to improve Gordon’s algorithm for application to problematical images. This can be achieved by taking into account the theory of edge detection operators [8] and visual perception criteria [9].