Image Thresholding Using Split and Merge Techniques with Log-Normal Distribution

Segmentation is the process of clustering the image pixels so that each cluster represent region in the input image and all the pixels inside the cluster share the same visual characteristics. Segmentation is a fundamental process in most image processing application. There are many segmentation techniques based on different methods such as region based method, edge based method, classification based method and hybrid method. Thresholding is one of the most known image segmentation technique. Although it is simple, it is effective in finding an optimal threshold value for obtaining better segmentation quality. This study presented a novel method for image thresholding based on Split-merge technique applied on histogram of the image instead of the image itself. We use the Log-Normal distribution to estimate the threshold value between every two successive classes. The proposed method is applied on artificial histograms and artificial images. A good segmentation result comparing to the existing methods is obtained. This is confirmed by experimental result