To suppress occurrence of noise in a reduced image during down-decoding.

An IDCT part 11 inputs a DCT coefficient after inverse quantization, and generates an IDCT output f1 obtained by reducing a coefficient of a high frequency component and an IDCT output f2 including the coefficient of the high frequency component, a motion compensation part 12 performs motion compensation by respectively using the IDCT output f1 and the IDCT output f2, and generates decoded imaged m1 and m2, a determining part 13 determines the possibility of generating noise during reducing the images from the decoded images m1 and m2, and a selecting part 14 selects a decoded image to be a reduction object when writing the decoded image in a memory from between the decoded images m1 and m2 on the basis of a determination result by the determining part 13.