Progressive image compression
T.D. Gedeon, Dominique Harris · 2003
In many applications of neural networks for image compression the main consideration is the decompressed image quality. The authors generally assume a feedforward network of three layers of processing units. All connections are from units in one level to the subsequent one, with no lateral, backward or multilayer connections. Each unit has a simple weighted connection from each unit in the layer above. The hidden layer consists of fewer units than the input layer, thus compressing the image. The output layer is the same size as the input layer, and is used to recover the compressed image. They can guarantee a consistent level of functionality of units in the compression layer based on their distinctiveness, and can progressively reduce the size of the compression layer for the desired level of image quality.>