Avoiding Redundant Processing in Gradient Based Edge Detection

nique can produce computational savings of up to fifty percent in some instances. Gradient based edge detection produces edge location information by convolving an image with a 11. THEORY kernel to calculate local intensity gradients. In many images, the area of the image that contains edge lo- The goal of image is decation information is small relative to the area of termine what is where. This is true at all levels of the image as a whole. During edge detection areas image analysis, including image segmentation where all areas of an image are processed. This leads to one wishes to classify all the points in the image into the redundant processing of areas that contain no regions. In an image the degree of certainty that edge information. In this paper an algorithm is de- be attached to what region a point belongs '0 scribed that combines two different approaches to is called the class resolution. Where that point is is image segmentation. ~h~ goal of the algorithm is the spatial resolution. Unfortunately these two are to identify areas of the image that do not contain incompatible. edges. hi^ information may then be used to drive The algorithm described in this paper eliminates selective edge detection which would avoid redun- redundant processing during edge detection. Two dant processing. different types of segmentation are combined to do this. Firstly an initial segmentation is carried out