A Novel and Locally Adaptive Approach to Correlating Multiple IR Sensors Using Partial Frames

Correlating multiple IR sensors is challenging because IR images may have different sizes, different poses, different illuminations, and different translations. Moreover, the IR images may not be perfectly aligned even after image registration. We propose a novel and high performance system for correlating multiple IR sensors using partial frames. First, for one partial frame containing a region of interest (ROI), we propose a sparsity based approach that uses this partial frame to search similar regions in other sensors. We have applied this locally adaptive approach in several applications, including face recognition, target detection in hyperspectral images, and region classification in hyperspectral images. For face recognition, the Yale B face image database, which contains face images with different illuminations, was used to evaluate our algorithm. We achieved close to 100% recognition rate. We can also handle rotated and scaled face images. Second, it is important to emphasize that our proposed correlation algorithm is parallelizable. In the past, we have implemented image processing algorithms using GPU and multicore CPUs. We have applied our fast parallel algorithms to speech processing, image registration, image inpaining of large images with high missing data rates, and genomic processing.