Real-Time Neural Network Based Color Classifier

This paper presents a real-time neural network based color classifier for machine vision applications. Classification is made by a self-organizing map and to achieve realtime performance, it is built a lookup table with all possible input colors and their corresponding classes. To minimize the size of the lookup table and to reduce the levels of input image quantization, it is applied an uniform quantization, an operation that reduces the total combination of possible colors and decreases the number of positions in the lookup table. By using the lookup table, our color classifier uses a single memory access strategy to mask a complex color classification operation saving processing time. Results of experiments performed are presented to show the performance of the proposed approach.