| Popis: |
Content based image retrieval is a technique which, by dint of a query image, can provide similar results to those eventuated by a query into a large images base. The traditional methods use the visual contents of the image; the requested image goes through a phase of extracting color, shape, or texture. The same goes for all of the images in the base. The images in question are then subjected to a direct comparison. In this work, we develop a content based image retrieval system using a method based on color features coded in the form of strings and genetic algorithms (color string coding and genetic algorithms). Herein, the problem lies in the selection of similar strings by means of meta-heuristic algorithms and the search methods based on the evolutionary concept using natural operations: selection, crossover and mutation. The aim is to select the fittest individuals in a given large population. Afterwards, we compare the results obtained by modifying at intervals the size of the images to assess the influence of the size on our method, and therefore, to decrease the computation time. This approach purports to reduce the costs of extracting other features and to increase the retrieval precision. Finally, our system is evaluated against the other systems based on two measures: recall and precision. |
