MRFBased Active Contour Model 831 Introduction

As discussed in section 8.1, most segmentation algorithms can perform well only on certain types of practical images because of the applicability limitation of each modeling or ease of use. In this section, we will introduce a flexible and powerful framework for general-purpose image segmentation. In the course of our investigation, we use the following assumption: A successful segmentation is an optimal local contour detection based on an accurate global understanding of the whole image. This assumption stems from the fact that the global information of an image is generally crucial in local object identification, automatic searching initialization, and energy optimization. Therefore, we focused our work in three parts:

(i) Region segmentation of the whole image: This provides a reliable basis for decision-making and subsequent processing.

(ii) Local object boundary tracking: This will optimally fine-tune the contour of the desired object region.

(iii) Flexible identification mechanism: This will bridge parts (i) and (ii) systematically and also be extendable to allow additional control functions or prior knowledge.

Our global-to-local processing logic is similar to the design concept of mul-tiscale image segmentation algorithms [60]; however, subsequent processing is radically different. The multiple-scale based techniques use various processing methods applied to assorted resolutions of original images, normally from rough to fine. In the proposed framework, different models are employed to describe original image in different processing stages, from region based to pixel based and from global to local. Therefore, it is a hybrid solution by integrating the advantages of more than one algorithm.

In section 8.2 QHCF is employed as a deterministic implementation of MRF-based image segmentation. In this section, we use MRF as the basic model for the global region partition. To further track and enhance an object's boundary, we adopt the enhanced version of the active contour model (ACM) and minimum path approach (MPA) as the basic tracking method. A new scheme will be introduced to find ACM initial points based on the MRF region segmentation results so that it can automatically provide a reliable initialization. More specifically, unlike looking for those points in a potential field in previous solutions, we pick up the most reliable contour pixels from boundary points found by the QHCF and use the two-end-point based MPA to find the curves between every two adjacent ones. Then the whole boundary of object can be found by linking all these curve sections.

0 0

Post a comment