An image processing device includes: an acquisition unit that acquires a probability distribution of luminance of a pixel in a plurality of perspective images including the luminance of the pixel which is superposition of luminance components resulting from respective objects, based on an occurrence frequency of a perspective image in which the luminance of the pixel has respective values of a plurality of values； a distribution estimation unit that estimates a mixture distribution indicating the acquired probability distribution； a component estimation unit that estimates a component corresponding to a smallest expectation value distribution of which an expectation value is the smallest among a plurality of partial distributions that form the estimated mixture distribution, and forming at least a portion of the luminance of the pixel； a removal unit that removes the estimated component from the luminance of the pixel included in the perspective image.
