A novel unconstrained correlation filter and its application in face recognition
IScIDE'12 Proceedings of the third Sino-foreign-interchange conference on Intelligent Science and Intelligent Data Engineering
| Hi-index | 0.00 |
Correlation filters are a standard way to solve many problems in signal processing, image processing, and computer vision. This research introduces two new filter training techniques, called Average of Synthetic Exact Filters (ASEF) and Minimum Output Sum of Squared Error (MOSSE), which have produced filters that perform well on many object detection problems. Typically, correlation filters are created by cropping templates out of training images; however, these templates fail to adequately discriminate between targets and background in difficult detection scenarios. More advanced methods such as Synthetic Discriminant Functions (SDF), Minimum Average Correlation Energy (MACE), Unconstrained Minimum Average Correlation Energy (UMACE), and Optimal Tradeoff Filters (OTF) improve performance by controlling the response of the correlation peak, but they only loosely control the effect of the filters on the rest of the image. This research introduces a new approach to correlation filter training, which considers the entire image to image mapping known as cross-correlation. ASEF and MOSSE find filters that optimally map the input training images to user specified outputs. The goal is to produce strong correlation peaks for targets while suppressing the responses to background. Results in eye localization, person detection, and visual tracking indicate that these new filters outperform other advanced correlation filter training methods and even produce better results than much more complicated non-filter algorithms.