Image Enhancement And Analysis Algorithms Using Color And Texture Features

We propose a new method of enhancing contrast of color images based on human visual system. In this method we convert the RGB (Red, Green, and Blue) values of each pixel of any segment of the original image to HSV (Hue, Saturation, and Value) values. Then we segment the V component of the original image into the dark and bright parts using k-means image segmentation technique. Next we apply (again to each segment:dark and bright) the wavelet transform to the luminance value V component of the color image to get the approximate component which is converted by applying grey-level contrast enhancement technique based on human visual system. Then, inverse Wavelet transform is performed on the converted coefficients so that the enhanced V values are obtained. The Saturation components are enhanced by histogram equalization. The H components are not changed, because changes in the H components could degrade the color balance between the HSV components. The enhanced S and V together with H are converted back to RGB values. Our new method has effectively achieved a successful enhancement of any color images considering their darkness or their low contrast by taking any image as a whole and then dividing it into its dark and bright segments.We also achieve an optimal approach of textures analysis and classification by combining Wavelet Transform and Neural Network. To reach a suitable way for textures recognition we first use Wavelet Transform to decompose texture into sub-images which are in turn analyzed and finally features are extracted. The Neural Network uses the extracted features to classify the different types of textures. Here, we have analyzed five types of textures and for each five different pictures have been used. We have obtained more accurate results.We have also reviewed the main approaches of partitioning an image into regions by using gray values in order to reach a correct interpretation of the image. We mainly compare the region-based segmentation with the boundary estimation using edge detection. Image segmentation is an important step for many image processing and computer vision algorithms while an edge can be described informally as the boundary between adjacent parts of an image. A formal definition is elusive, but edge detection is nonetheless a useful and ubiquitous image processing task. After comparing we have come to a conclusion that the edge detection has advantage of not necessarily needing closed boundaries and also its computation is based on difference. The region-segmentation in spite of improving multi-spectral images has the drawback of being applied only on closed boundaries. To reach the result of edge detection we have used the technique of performance metrics and canny edge detection. We have applied canny ground truth to acquire more features via displaying more details.