CBIR Using KEVR Vector Quantization applied on Gradient Mask Edge Images

The paper presents image retrieval technique based on shape features extracted with the help of seven gradient masks (Robert, Sobel, Prewitt, Canny, Laplace, Frei-Chen and Kirsch) and Kekres Error Vector Rotation (KEVR) vector quantization codebook generation method technique. First shape features are extracted from image of the database using various gradient masks and slope magnitude method, to get edge images. Then Vector Quantization codebook generation algorithm (KEVR) is applied on the obtained edge images, which extracts the shape texture features. Here seven assorted codebook sizes (8, 16, 32, 64, 128, 256 & 512) are considered with seven different gradient masks resulting into 49 variation of the proposed method. This method of image retrieval is applied augmented Wang image database on 1000 images. The database consists of 11 categories of images. Five images from each category are taken as query to find precision and recall values for CBIR. The crossover point precision, and recall values is considered for performance evaluation of all proposed variations. The image retrieval using canny gradient mask with slope magnitude method and KEVR has given better performance for codebook of size 512