Study On The Region Of Interest And Sparse Representation Of The Human Face Super-resolution Reconstruction

As people continue to improve the image quality requirements, image processingtechnology has become popular nowadays research areas. Image super-resolutionreconstruction refers to a low-resolution image by the one or more images through imagedegradation sequence, one or more images to reconstruct high-resolution images of theprocess. High-resolution images are represented by units of pixels smaller scene in reality,reflecting greater image detail, can provide richer information. Through the signalestimation theory, not only a good solution to the problem of the sensor array density limit,while the imaging optical system degradation problems and has a good deal. Currently, thesuper-resolution reconstruction technique has been widely used in military investigators,medical imaging, public safety fields.This important part of the contents through the three proposed and studied based onthe region of interest and sparse representation of the human face super-resolutionreconstruction:First, the region of interest on the face feature extraction research, analysis of threeclassic facial region of interest algorithm low-resolution face image feature extractionaccuracy, after experimentally derived Active Appearance Model algorithm at a resolutionof low-resolution facial image recognition accuracy is much better than other algorithms.Then, through the sparse representation based super-resolution reconstruction ofresearch, explains the basic theory of sparse representation, sparse coding algorithmoptimization, algorithm design updated dictionary; and learning through the jointsuper-resolution dictionary constructed using sparse representation methods forlow-resolution images of the super-resolution reconstruction, resulting in high-resolutionimages. Also compared Bicubic interpolation algorithm and the gap between the peaksignal to noise ratio of the reconstruction algorithm for experimental research results showthat the dictionary size and the number of samples have a direct impact on how muchsuper-resolution reconstruction after the results.Finally, we propose a region of interest based on sparse representation of the humanface and super-resolution reconstruction algorithm process, but also an optimizedhigh-resolution dictionary method, which makes high-resolution dictionary fully reflectsthe low resolution rate the performance characteristics of the image can be more efficientand accurate for the dictionary matching. Meanwhile verified by experiment, to learnmore dictionary for training can improve the quality of the reconstructed image, peopleface image more prominent part of the region of interest, more recognizable recognizable.