| Resumen |
In this paper we describe and test three different methods for image retrieval. The three proposals use exemplar images as inputs and display the most similar images. The three methods make use of the well-known Daubechies 4 wavelet transform. In all cases the images are of 256x256 pixels. First method: Each image to be further retrieved is first represented by a vector of 48 coefficients of corresponding wavelet decomposition of the image, 16 of the red, 16 of the green and 16 of the blue of the RGB color space. During image retrieval, a query image is presented to the system. It is decomposed again into a 48-sized vector. This vector is compared against all the vectors representing the N images already organized. The n, with n<<N, most similar images are displayed. The Euclidean distance was used for comparison. Second method: To each image the histogram of the circular image inscribed in image is first computed. This histogram is then represented by a vector of 48 coefficients of the corresponding wavelet decomposition of the histogram. The same retrieving idea adopted for first method is here used. Third method: Each image is first decomposed into 16 sub-images (4 by 4 each one). To each of these sub-images its corresponding histogram is then computed. To each one of these histograms just one coefficient is computed using the same wavelet decomposition. The 16 coefficients (one for each sub-image) are integrated in a vector, 16 of the red, 16 of the green and 16 of the bl |
