To effectively convert images by reducing the amount of data to be prepared in the case of performing expansion to the images of various sizes based on the compression data of the images and shortening the processing time for inverse quantization in the case of generating reduction images.

A decoding part 1 Huffman decodes and run length decodes the compression data of one block and generates quantized data Ruv. The quantized data Ruv form the matrix of 8×8 for instance. Since Huffman encoding and run length encoding are reversible encoding, the decoded quantized data Ruv are completely the same as the quantized data Ruu at the time of performing JPEG(joint photographic expert group) compression. A reduction part 2 omits the high frequency component of the quantized data Ruv and performs conversion to the matrix of a small dimension. Thus, the amount of the data to be prepared is reduced in the case of performing the expansion to the images of the various sizes based on the compression data of the images. Then, in the case of generating the reduction images, the processing time for the inverse quantization is shortened and the reduction images faithful to source images are obtained.