1 3-3 1 A Competitive Learning Algorithm for Color-Based Image Retrieval

The paper aims to build up a common color codebook, which represents well the database color information under consideration to improve the performance of the image retrieval problem. The frequency sensitive competitive learning neural network algorithm is used for this task, in order to form the set of prototypes that represents well the environment. The similarity measure is a performed aid of a suitable distance measure. An image database of 250 images is used to test the performance of the algorithm. Generally. content based image retrieval need to automatically extract primitive visual features from images and to retrieve them on the basis of these features. Humans use color, shape, and texture to understand and recollect image contents. Therefore, it's natural to use these features for the automatic image retrieval application. Image retrieval, with color as feature for image matching (4), tries to obtain a list of images from the database which are similar in color to a given query image. The measure of similarity between feature values of two images is based on all most case on color histogram matching techmque. So we suggest in this paper to use a common color codebook. that contains prominent and distinctive colors, for the interpretation of the color information of the entire database. This suggestion is motivated by several obvious and objective remarks. First, the human eyes can not distinguish close colors very well, ;md the database contains a large set of similar colors. Second. The storage and the computational time required for the retrieval process can be reduced. llurd, the database can consist of certain kind of images, for example rural or urban images, where certain colors are dominant. So. it's very useful to use a learning algorithm that can extract pertinent color information. In tlus paper, we use a data reduction technique. based on a competitive learning neural network algorithm to build up the color codebook. Figure 1 shows the image retrieval model.