Object Reconstruction In Image Based On Compressed Sensing

The recently developed new theory compressive sensing (CS) can ensure that signals can be sampled in the rate much lower than that Nyquist sampling theorem required and the sampled signals can be reconstructed exactly in the case of no loss of signal. The theory can reduce the sampling frequency and the price for data storage and transmission largely, and cut down the time and cost for signal processing and calculation.In this paper we introduce the application of CS in image processing. CS and some methods in image processing are combined to achieve our goal, reconstruction of image object. The CS theory is used to reconstruct object in images, and the detailed efforts are made as follows:(1) When CS is used in the processing of natural images, cutting original image into blocks can ensure the quantity of reconstruction as well as saving memories and improving the processing speed. Without projection to another space, the images can be reconstructed by minimum mean square error (MMSE) linear estimation which can achieve a better result than L2-norm.(2) Combining CS with data similarity. The blocks of detected image are divided into two parts according to an empirical threshold by comparing the similarity and reconstructed by two different strategies respectively, which can ensure that the object is reconstructed better than background.(3) Combining CS with visual attention model to obtain the saliency map of an image and reconstruct the image. The saliency map illustrates the regions of interest in an image, which is the weight matrix corresponding to the original image. The weight matrix is used to weight the measurement matrix to get weighted measurement vectors with information of object only.