Abstarct: Image enhancement is one of the most important preprocessing operations especially in medical and astronomical imaging and some other applications. Due to technical limitations and deficiencies, some of imaging systems used to take, print or display the pictures, apply non-uniform changes on the images which lead to a reduction in quality. This artifact is called the “effect of gamma” and the recovery process is named “gamma correction” respectively. As the imaging systems are not fully able to represent the true color, depth and texture of the objects, the gamma effect is not equal for different parts of the image. In order to recover the real scene, the correction process should be adaptive and estimate the gamma value in a local way. In the present thesis, we present an adaptive gamma correction approach for image enhancement. Here image enhancement process affects the image`s brightness, contrast and the details. We propose three methods for this purpose. Most of existing gamma correction methods apply a uniform gamma correction across the image. Considering the fact that gamma variation for a single image is actually nonlinear, the recommended approaches in this thesis consider local and adaptive gamma correction. The first method uses windowing technique and texture features extracted by co-occurrence matrix to determine the appropriate gamma value for different parts of the image. In order to get local estimations of gamma value, the image is divided to overlapping windows with 30 different gammas (0.1 to 3) applied to each window. One of these gammas is the closest value to the real gamma and thus enhances the image. In chapter 5 it will be discussed that an image with varying intensity levels has a minimum homogeneity value which indicates the amount of image details. Using the homogeneity feature of the co-occurrence matrix to measure the amount of image details, a proper gamma value will be assigned to each window. In the next method, segmentation is used instead of windowing. Again the homogeneity feature of the co-occurrence matrix is used to define the gamma value for each segment. The third method uses SVM classifier for local estimation of gamma. At first, the training images are constructed from various standard images applying different gammas on them. After windowing each of the training images, a set of nine features that characterize the image contents are computed from its pixel intensity histogram, co-occurrence matrix, and discrete cosine transform (DCT) domain. Along with the applied gamma value, these features are fed to SVM to estimate gamma values of each associated window of new image. Various experiments were done on natural and medical images to testify the applicability of our proposed methods. In this thesis, subjectve and objective measures are used to evaluate the efficiency of the proposed methods. The results confirm that these methods outperform other common approaches.