F. Roli and S. Vitulano (Eds.): ICIAP 2005, LNCS 3617, pp. 312 – 319, 2005. © Springer-Verlag Berlin Heidelberg 2005 Enhancement of Noisy Images with Sliding Discrete Cosine Transform Vitaly Kober 1 and Erika Margarita Ramos Michel 2 1 Department of Computer Science, CICESE, Ensenada, B.C. , Mexico vkober@cicese.mx 2 University of Colima, Colima, Mexico ramem@ucol.mx Abstract. Enhancement of noisy images using a sliding discrete cosine transform (DCT) is proposed. A minimum mean-square error estimator in the domain of a sliding DCT for noise removal is derived. This estimator is based on a fast inverse sliding DCT transform. Local contrast enhancement is performed by nonlinear modification of denoised local DCT coefficients. To provide image processing in real time, a fast recursive algorithm for computing the sliding DCT is utilized. The algorithm is based on a recursive relationship between three subsequent local DCT spectra. Computer simulation results using a real image are provided and discussed. 1 Introduction Many different image enhancement techniques have been introduced to improve the visual appearance of images [1-7]. These techniques may be broadly divided in two classes. The first class is based on decomposing an image onto high- and low- frequency signals, manipulating them separately and then combining them. Examples of such methods are homomorphic filtering [1] and unsharp masking [2]. The second class consists of various histogram modification techniques [3]. The classical unsharp masking is one of the most commonly used methods for image enhancement because it works well in many real applications. In this method a fraction of the high- frequency signal of an image is added to the original image itself to form a locally enhanced image. Drawbacks of the unsharp masking are as follows. A linear highpass filter makes the system very sensitive to noise. This results in undesirable noise enhancement in flat and high-contrast areas of even slightly noisy images. The operation also uses a constant scaling factor that may lead to overshoot artifacts in high-contract areas of the image. Various methods have been proposed to improve the performance of the unsharp masking [4-6]. The use of quadratic filters instead of a linear highpass filter enhances details and edges in accordance with a human perceptual criterion. These filters can be described as local mean weighted highpass filters. Weighting the highpass filter output by the local mean value leads to enhancement in dark areas less than that of in bright areas. This coincides with Weber’s law, which states that the just noticeable brightness difference is proportional to average background brightness. Consequently the perceived noise is reduced