The paper studies two methods for detecting and removing channel error patterns in images transmitted by two-dimensional differential-pulse-code-modulation (2-D DCM) with fixed-length words. The methods are based on statistical classification and filtering procedures. For the predictions A+(C-B)/2 and A+(D-B)/2 in combination with 4-b DPCM, it is demonstrated experimentally by computer simulation that the effect of channel errors can effectively be reduced at bit-error rates (BERs) up to approximately 0.1%.