With a video compression device configured to compress one pixel per cycle, a predictive pixel generation unit generates predictive pixel values of a plurality of predictive modes defined assuming local decode pixels read from predetermined positions of a line memory as upper reference pixels and an input original image pixel positioned on the left side of a pixel to be compressed as a left reference pixel for the pixel to be compressed. A predictive mode determination unit calculates a predictive error in a unit based on differential values between the pixel value to be compressed and the predictive pixel value, and selects a minimum predictive mode. A DPCM unit generates a minimum differential value between the minimum predictive pixel value and the pixel to be compressed assuming local decode pixels as upper reference pixels and a local decode pixel value one cycle before as a left reference pixel.