Histogram Oriented Gradients and Map Seeking Circuits Pattern Recognition with Compressed Imagery

Object recognition from compressed imagery has the potential of saving storage and processing time. However, the selection of the compression technique needs to be optimal for the pattern recognition algorithm's feature extraction. We consider here image compression in the presence of Map Seeking Circuits (MSC)-based object detection, a technique which relies on finding the edges within the imagery and correlating the patterns with the object of interest. Our results indicate that for JPEG and JPEG 2000 compression, high compression ratios do not affect object detection and localization results to any significant degree when using MSC. In addition, we also consider the effect of compression on Histogram Oriented Gradients (HOG)-based feature extraction, finding again that the effect is manageable.