Efficient and robust level set model for extracting regions of interest in X-ray welding images and MRI brain images

Extraction of the region of interest (ROI) from the X-ray welding images and MRI brain images is extremely challenging due to their poor quality, low contrast, high noise and blurry boundaries. The level set technique is one of the most often employed in this field, but its main problem is the initial contour selection. To solve this problem and obtain reliable and accurate extraction of ROI, we propose in this work an efficient and robust level set model (ERLSM). More specifically, our model is a hybrid between the conventional level set method (CLS) and the possibilistic c-means clustering (PCMC) method. The main idea of PCMC is to initially divide the input image into clusters. Then, the obtained results are used to initialize and estimate the controlling parameters of the CLS method. Experimental results on X-ray welding images and MRI brain images show that the ERLSM model offers very good performance and confirms its efficiency and robustness against image noise. Moreover, the proposed model has the advantage of high computational speed under different processing conditions. It is able to extract ROIs with high accuracy after very few iterations and eliminates the sensitivity problem of CLS to the initial contour.