Application Of Steerable Filters In Image Edge Detection

Edge detection is a common task in Image Engineering. The directional derivative filter is useful in edge detection. Classic grads operator is composed with directional derivative filters on two directions. It is often necessary to rotate a filter to different angles in applications as feature extraction and human face reorganization. Steerable filter is designed for this. Steerable filters are usually generated according to one or a set of known filters, this process is called rotation of a filter. It is required that the rotated filter be consistent with the original filter, namely the direction of the filter is changed while other characters remain unchanged. Filters on the required directions can be manually defined in simple tasks, this certainly limits the available directions to several one, and the consistency is not assured. It is more reliable to set up a arithmetic to ensure the consistency of the rotated filters. Most of these arithmetic can rotate a filter to any angle provided that the filter fulfills all requirements of the arithmetic. The requirements of traditional arithmetic are usually strict.This paper explores a model of steerable filter to simplify the calculation of filter rotation. The definition and calculation of directional derivative on a grid image is the foundation of this research. Illuminated by the process of calculating grads with classical grads operators, the concept of cruciate pixel group is provided. A cruciate pixel group is a basic unit for the calculation of the directional derivative, and a directional derivative filter can be described as a combination of several cruciate pixel groups. Defining variable weights for each cruciate pixel group enables the model to fit different filters, and the function of the weights contains the characteristics of different filters. Experimental result verifies the effectiveness of the model in the rotation and analyzing of the filters.