Freak Descriptor With Spatial Pyramid Kernel For Scene Categorization
Qiong Yao, Xiang Dong Xu · 2015
Spatial Pyramid Kernel performs good on challenging scene categorization tasks for combining multi-resolution and spatial information of image features , while the popular feature extraction still be the global SIFT or GIST, neglecting more recent , effective and efficient descriptors, such as FREAK ,which is faster to compute , more compact , and with lower memory.We partition image into fixed sub-regions, compute dense freak descriptor over fixed subregions' center points instead of on interest points, then use the spatial pyramid kernel recognition method for scene categorization.We make two comparisons, the first is between the FREAK (Fast Retina Keypoint) and the SIFT descriptor under the same scene categorization framework and the second is compute dense freak descriptors over fixed subregions' center points versus on interest points.The results show that the dense freak descriptor ensures a very competitive categorization performance with lower computational cost and less memory .It indicates that the FREAK descriptor is avaiable alternative to the SIFT descriptor for the problem of scene categorization and particularly useful for real-time-recognition systems for low-resource devices such as mobile phones.