A SMOOTHED DPMVL FOR INTERACTIVE IMAGE SEGMENTATION AND ENHANCED ADAPTIVE MRF FOR SEGMENTATION REFINEMENT

G.S. Gowri; Dr. P.Ponmuthuramalingam

Volume : VII, Issue : IV, April - 2018

A SMOOTHED DPMVL FOR INTERACTIVE IMAGE SEGMENTATION AND ENHANCED ADAPTIVE MRF FOR SEGMENTATION REFINEMENT

G. S. Gowri, Dr. P. Ponmuthuramalingam

Abstract :

Image segmentation is a fundamental step in many areas of computer vision including object recognition, video surveillance, face recognition, fingerprint recognition etc. It provides additional information about the contents of an image by identifying edges and regions of similar color and texture. Although a first step in high level computer vision tasks, there are many challenges to ideal image segmentation. Interactive image segmentation is a way to extract foreground objects in complex scenes using simple user interaction. The key to success in interactive image segmentation is to preserve characteristics of the user’s interactive information and maintain global data effectively.In Smoothed Dirichlet Process Multiple view learning (sDPMVL) with improved adaptive Markov Random Field (MRF) Model utilized to more accurate and efficient segmentation than traditional MRF model. However the parameter selection is the critical issues in this approach. To overcome this issue the sDPMVL–adaptive MRF with Modified Graph cut utilized to achieve the efficient segmentation and smoothness in the segment label.However, the shape information of objects was not considered in these approaches which this leads to decrease the smoothness of the segment labels. To further increase the smoothness of this approach the sDPMVL– adaptive MRF with Modified Graph cut and Adaptive shape prior utilizes to achieve efficient smoothness.In these approaches the HMC (Hamiltonian Monte Carlo) algorithm utilized to learning new parameter and inference. However, convergence rate of Markov chain and the decorrelation time between independent samples can be problematic. This causes to degrade the accuracy of smoothening and segmentation in the segment label. In order to overcome this issue in this paper the Simulated Annealing (SA) based Smoothed DPMVL (sDPMVL) Adaptive MRF with Modified Graph cut and Adaptive shape prior proposed to improve the new parameter learning and inference.The Performance of the proposed approach is analyzed through accuracy, precision and recall.