This paper proposes a new chrominance feature extraction method in HMMD color space. Image dependent multi-level thresholding is performed in the HMMD color space to obtain the 64-IeveI quantized images. The occurrence count of each color pixel represents the color information of those quantized images. This technique is tested over Wang's database of 10 different category images. The distance measure of this feature between the query and database image are calculated. Then, the proposed method performance is evaluated using average precision and recall. Moreover, the proposed method is a benchmark against the state-of-the-art color feature extraction methods and gives approximately 6.3% to 18.05% and 7.54% to 14.52 % high precision and recall than the conventional techniques.