An image compressor of the present invention includes a difference calculating unit and a difference requantizing unit. The difference calculating unit calculates a predicted difference between a pixel value of image data and a predicted value (a calculated value of the pixel value predicted from known data). When a magnitude of the predicted difference exceeds a predefined threshold value, the difference requantizing unit applies the coarse quantization steps to requantize the predicted difference to keep a bit range to predetermined size. In this manner, compressed data, in which a bit length of the predicted difference for one pixel is adjusted to a fixed bit length, is generated.