Multi-scale feature identification using evolution strategies
Image and Vision Computing
| Hi-index | 0.00 |
This paper proposes a method for texture image database retrieval using the universal classification theory. The wavelet transform is taken for input images, and wavelet transform coefficients in each subband are further processed to obtain the type-based discrimination measure between images. The type-based empirical sequence classification rule is asymptotically optimal. Simulations show that the type-based retrieval scheme is capable of yielding very good texture retrieval performance. Compared with the conventional subband energy-based distance measure for wavelets coefficients of images, our method yields on average about 6% to 8% higher texture retrieval rate for an image database of 97 textures. The superior distance measure we present can be used for image classification as well as image retrieval.