A motion-setting section (61) sets a motion amount and a motion direction for obtaining processing coefficients. A student-image-generating section (62) generates student images obtained by adding a motion blur to a teacher image not only based on the set motion amount and the set motion direction but also by changing at least one of the motion amount and motion direction in a specific ratio and student images obtained by adding no motion blur to the teacher image. A prediction-tap-extracting section (64) extracts, in order to extract a main term that mainly contains component of the target pixel, at least a pixel value of pixel in the student image whose space position roughly agrees with space position of the target pixel in the teacher image. A processing-coefficient-generating section (65) generates processing coefficients for predicting the target pixels in the teacher images from the pixel values of extracted pixels based on a relationship between the pixels thus extracted and the target pixels in the teacher images. The processing coefficients that are suitable for any motion blur removing which is robust against any shift of the motion vector can be generated through learning.