An image compression apparatus includes: a line memory unit for holding pixel values for at least one immediately previous line in the lines to be processed of an image to be compressed; a division unit for dividing pixels of the line to be processed into 2n-pixel blocks; a first estimation value calculation unit for extrapolation-estimating the 2n-1-th pixel and the 2n-th pixel in the blocks divided, and obtaining an estimated value in processing for the first hierarchical level; a second estimation value calculation unit for interpolation-estimating certain pixels and obtaining an estimated value in the processing for the second hierarchical level through the processing for the n-th hierarchical level; and an estimation coding unit for obtaining an estimation error from the estimated value, converting the estimation error into a quantization number, converting the quantization number into a variable length code, and obtaining a compression code.