An edge fusion approach based on the concept of multiple detector behavior

One of the fundamental steps in image processing is edge detection. The performance of any edge detection method is highly affected by the choice of parameters, e.g. thresholds. For some applications, such as identification of objects in a scene, the interesting regions do not usually cover the whole image. Hence, applying edge detection with global parameters to these images will detect edges of interest (EoI) as well as edges caused by background details, illumination or reactance. We present here a new edge fusion approach that aims at identifying edges of interest and discarding unwanted edges. For this purpose, different parameter values are used to generate edge images from a given gray-scale image. The image is then divided into small windows and for each window the local detection accuracy of the various parameter values are estimated. Finally, the edge images are fused to produce a new edge image by locally choosing parameter values that achieve maximum detection accuracy. Initial results have shown that this method has the potential to produce very promising results.