To obtain a decoded image nearly the same as an original image with a small data amount.

Correction data are outputted after correcting the compressed data which are obtained by thinning out the original image in a compression 21. The original image is predicted on the basis of the correction data in a local decoder 22. Then, the predicted value of the original image is outputted. The prediction error of the predicted value to the original image is calculated in an error calculator 23. The above processing is repeatedly executed in a determination part 24 until it is determined that the prediction error is below a prescribed value. The correction data when the prediction error is below the prescribed value are outputted as the coding result of the original image. The device and method for coding an image can be applied, for example, to an image coder.