Image Segmentation using Learning Vector Quantization of Artificial Neural Network

Abstract

This paper presents color image segmentation application for road-sign detection and recognition system based on Learning vector Quantization (LVQ). In traffic-sign maintenance and in a visual driver assistance system, road-sign detection and recognition are two of the most important functions. Our system is able to detect and recognize the signs. Road signs provide drivers important information and help them to drive more safely and more easily by guiding and warning them and thus regulating their actions. The RGB images are converted into HSV color space, and segmented using LVQ depending on the hue values of each pixel in the HSV color space. LVQ neural network is used to detect the color from the images and recognize them with high accuracy and speed. The detection and recognition of traffic signs represents an important issue in the field of traffic sign analysis. The paper researches the existing traffic sign detection and recognition methods. Special attention has been given to the methods based on artificial neural networks. Some techniques extracted various features from the images containing traffic signs and used various detectors and classifiers to perform detection and recognition. The following examples of these methods can be mentioned: Histogram of Oriented Gradient (2) or (1) and (6) or Support Vector Machines (SVM) (5). Some authors have also used artificial neural networks (4). This paper, however, uses a method of neural network (NN) which has given successful results. For individual object identification in humans, Color discrimination plays an important role. Image segmentation is the basic and first step regarding image analysis and pattern recognition. In image processing, image segmentation is not only the critical and essential component but also it is a very difficult task. The actual operation of the algorithm determines the accuracy and quality of image analysis. Color segmentation is a process of extracting one or more connected regions from the image domain. These connected regions are satisfying the uniformity (homogeneity) criterion which is derived from the spectral components. These components are then defined with respect to some color spaced model such as RGB model. RGB is the most common model. In RGB model, a color point is defined by the color components of the corresponding pixel which are defined in the form of Red (R), Green (G) and Blue (B). This paper considers a color image segmentation problem as a pixel classification problem. .