Measuring the effective coverage of the image databases

The paper presents new methods for estimating effective coverage of the image databases in spatial information (SI) vs. colorfulness (CF) space. Existing and commonly used metrics are vulnerable to databases containing very different images in terms of SI and CF, which can maximize the relative ranges, while preserving, in specific cases, high uniformity. The area covered by convex hull created on SI × CF space can be large, but the interior of the convex envelope might be poorly filled. To overcome this flaw, a fill factor is proposed, which determines how well the convex hull area is utilized. In the paper the method is described in details and presented on a few existing databases, as well as on artificial, non-existing databases to prove its robustness in different scenarios. Provided results show that new metric is closer to the actual coverage than existing methods.