Content-based image retrieval using new color histogram

This study has proposed a new method of color representation, and a method of similarity measurement in order to overcome the disadvantages of a color histogram. The existing color histogram intersection method uses only the frequency value of the same color, after color quantization; which causes quantization errors. To reduce this error, it calculated the mean value of RGB color components and color frequency in each color region, selected them as the representative value of the similar region of a relevant color, then stored this in the DB as a feature vector, and finally, measured the similarity between color images by applying fuzzy theory. As a result, the color histogram has retrieved similarity between images more precisely than the existing method did. The study experimented on 1,000 color images by the new color histogram retrieval method, and found it more precise than the existing method.