Traffic sign segmentation using supervised distance based classifiers

Abstract

Image Segmentation is an important process in computer vision techniques. This paper presents a color segmentation algorithm based on Supervised Distance Classifiers. Noting that the traffic signs are usually represented by six main colors, the classifier is trained using random values representing the six main colors in HSV (Hue saturation and value) color space. In the testing phase, the image is converted to HSV images and each pixel in the HSV images is compared with the six colors group's prototypes using LVQ (linear vector quantization). The output consists of six segmented images representing the six different color regions. Color restoration technique (MSRCR) for image enhancement is applied to images with low light conditions. The technique has the ability to segment the images in the six color regions in various experiments done in different weather conditions with good accuracy and speed.