|
|
|
|
Keypoint Recognition with Histograms of Normalized Colors
Clark F. Olson and Siqi Zhang In Proceedings of the 13th Conference on Computer and Robot Vision, pages 311-318, June 2016. Download (6.14 MB) Keypoint recognition is an important component of many object recognition and image classification systems. However, the color information present in images is not well incorporated. Previous techniques mostly rely on derivatives that only use the rate of change, rather than the actual color, and/or neglect the spatial relationships between the colors. We describe a new keypoint descriptor constructed with normalized color histograms that preserves color information and spatial relationships while maintaining invariance to illumination brightness variation (bias and gain). When combined with shape information, our descriptor surpasses the performance of previous techniques. |
