In an image processor, image data of red, green and blue are converted to image data of lightness and chromaticities, and they are divided into pixel blocks. Then, image data of each block are compressed to data having 4 gradation levels smaller than 256 levels of the image data by generalized block truncation coding. Then a capacity of image memory can be decreased. An attribute of the image data is decided for each block according to the compressed data. The attribute can be corrected by taking attributes of adjacent blocks into account. Edition of image data such as edge emphasis can be performed according to the attributes. The compressed data are further compressed by deleting data which can be specified according to the attribute.