Here we proposed algorithms for CBIR system on the basis of texture, shape,and color based feature extraction and matching of color and texture.We used the Discrete Wavelet transform for decomposition of images and clusters calculations using modified K-Means clustering.We extract texture,shape, and color and finaly measure similarity between query image and database image and reduced semnatic gap between local features and global features. Integrated approach retrive more accurate image, reduce semantic gap between local and high level features.The time taken by Modified K-Means is less as comparison to other algorithms.This is more optimized method for small as well as large database.