Noise reduction is performed on the basis of characteristics of an image in a detection range. A noise reduction block 4′ performs a second-order differentiation process and a symmetry process to decide adjacent pixels with which noise reduction is preformed for an attention pixel. With the pixel level of the attention pixel in the detection range and the pixel levels of adjacent pixels used for noise reduction, an arithmetic mean processing section 16 calculates a mean value. A median filter 17 selects a median value. With the number of pixels used for noise reduction, it is determined whether the image in the detection range contains a flat portion, a ramp portion, or an edge. The mean value and the median value are weight-added with a weighted coefficient that are changed on the basis of characteristics of the image. The result is substituted for the level of the attention pixel. When the attention pixel is an isolated point, an all-pixel median filter section 31 selects a medium value of the levels of all the pixels in the detection range including the attention pixel and substitutes the median value for the level of the attention pixel.