A compact color descriptor for image retrieval

The resource usage in Content-Based Image Retrieval is a frequently neglected issue. This paper describes a novel compact feature vector based on image color histograms in the HSL color space. The images are represented using only 10 bytes per image. It is shown that, in the context of Query-by-Example (QbE) usage scenarios, the method described achieves retrieval performance close to the state of the art image retrieval methods that use considerably more memory. It is also shown that the described method outperforms other methods with similar memory usage.