Multi-resolution architecture of graded memory

An approach to storing static images in some predefined resolution and retrieving stored static images in multi-resolution using multilayer Hopfield neural network is proposed. Here, the Hopfield network is used as a memory, which stores images in predefined resolution. The neural network is trained to store certain number of images. During image retrieval from the network, following schemes are proposed. Firstly, down sampled version of the stored image is provided in some fashion as the query mage, the network initially gives out a coarse image. This coarse output image is processed and fed back as the input image to the memory again. The output image retrieved this time will be better than the one that was obtained initially. This is repeated and the output of the memory becomes better and better as the time progresses. Secondly, the network can be configured to render an image in such a way that the Region Of Interest (ROI) of the image becomes better and better over the period of time. In the third scheme, the network can be configured to render an image with the ROI of the image becoming more blurred over the period of time. The above mentioned schemes have been simulated using MATLAB for the following four images Lena, Baboon, Barbara and Stefan. The results are shown both in the tabular and graphical form.