Local Orientation Coding and Adaptive Thresholding for Real Time Early Vision

This paper suggests a new preprocessing algorithm for grey level images. It is designed for computer vision tasks and based on local orientation coding. The method can be employed as an edge detector or as a means to label the orientation of neighborhoods in the imagery. It includes an algorithm to determine the necessary thresholds automatically. In addition to presenting the method an in depth analysis of the adaption mechanism is given. Due to the simplicity of the algorithms short execution times can be achieved. The methods were developed to preprocess grey scale images for a neural network based car detection and tracking system. The implementation meets the speciications for a real time system.