Improvement of Quadree-Based Region Segmentation

Concerning the domain of Image Processing, image segmentation step is considered to be a wider term as a necessary requirement for the progression of the essentiel data out of the given image. Therefore the image segmentation is a very important step in most computer vision systems. Our aim is to improve the quality of the segmented image by the Quadtree division method, for this we proposed three approaches: the first is to segment the palette of gray levels in 32 intervals to reduce the blocks number obtained in order to avoid over-segmentation, the second is to cooperate two methods one for the regions-based segmentation using the Quadtree method and the other for the edges-based segmentation using Sobel operator or Canny operator, in order to obtain reliable result by exploiting the advantages of both types of segmentation. The third and last is a hybridization between the Quadtree and one of the segmentation by clustering method that is K-means, using the peaks number detected by the equalized histogram. For a good automation of the system, we have assigned this peaks number to K the class number in Kmeans.