Threshold modeling for cellular logic array processing based edge detection algorithm

Edge detection is one of the basic methods for various image processing functions such as image analysis, image segmentation, pattern recognition etc. This is a process to find out discontinuity of intensity in image. If some or all neighboring pixels form a convex region of same gray level intensity, then there exists an edge. In order to distinguish between different level of intensities in edge detection, a threshold is required which is usually different for different type of images due to variation in level of intensities. This paper proposes and compares two methods namely global and local thresholding to model the value of threshold through quantitative empirical method for cellular logic array processing based edge detection method. The performance of the modeled algorithms is measured by F1-score, recall-precision break-even-points and performance ratio. Experimental results show that the local thresholding approach gives slightly better F1-score and performance ratio for all scenarios of six Berkeley Segmentation Database images and respective ground truths. It has also been found that best percentage of threshold value can be determined in a better way by break-even-point rather than by best F1-score. The proposed approach reduces false edge detection and make threshold selection automatic for every scenario.