Image retrieval is defined as the process of retrieving same or similar of an image queried from a digital image database. Although a digital image is composed of pixels, the query is not performed at the pixel level but it is carried out at level of vectors representing digital images. In other words, it is computationally necessary to represent with vectors both queried image and images in database. The similarity between queried and database images is computed by vector operations. The representation of images by vectors is called feature extraction process and it is the most significant stage of content-based image retrieval (CBIR). Histograms of gray scale images are typical feature vectors. On the other hand, as there are three different channels in color images, the histograms which represent images are three-dimensional arrays, which will increase the computational cost of the system considerably. For this reason, researchers have preferred to use color quantization or reduce the number of colors in color images. The color reduction process is called as vector quantization, but it is not always possible to produce the same result. The reason is that the developed algorithms so far look for solutions with randomly generated color vectors initially. Linde-Buzo-Gray (LBG), K-means and fuzzy c-means algorithms are typical examples of such solution approaches. In this study, a new image retrieval method has been proposed by using the recursive mean-based color reduction approach. In the proposed strategy, firstly, averages were calculated from the histogram of each color channel and consequently multi-level thresholds were obtained. Using the thresholds obtained, RGB color space was sliced into sub-prisms. The pixels in the created sub-prisms were assigned to the same class and color reduction was made by using the means of pixels in the related class. One-dimensional histogram was obtained with the help of class indices and the number of pixels allocated to the related classes. In the last stage, the obtained class-based histogram was assigned as feature vector and content-based image retrieval was performed. The results were obtained with the proposed algorithm and LBG algorithm. Additionally, comparisons were made.
Primary Language | Turkish |
---|---|
Subjects | Engineering |
Journal Section | Articles |
Authors | |
Publication Date | January 31, 2020 |
Published in Issue | Year 2020 |