A compression device using a prediction coding method performs a prediction for one line in an image to be compressed by using each of a plurality of the same types of predictors as plural types of predictors, and determines one quantizer out of plural types of quantizers based on a prediction error amount of a predictor having the smallest prediction error. It also performs a prediction for a prescribed pixel block in the one line, and identifies one predictor out of the plural types of predictors based on a type of a predictor having the smallest prediction error. Then, it compresses data of the prescribed pixel block by using the identified predictor and the determined quantizer. A decompression device decompresses the compressed image data to generate the image to be compressed by using an inverse quantizer corresponding to the determined quantizer and the identified predictor.