An endoscope image processing device includes a processor. The processor sequentially receives an observation image obtained by performing image pickup of an object inside a tubular portion of a subject, performs processing for detecting a region of interest with respect to the observation image, performs judgement processing as to whether degradation of visibility of the region of interest included in the observation image is predicted, and performs emphasis processing for emphasizing the position of the region of interest when a judgement result indicating that the degradation of visibility of the region of interest included in the observation image is predicted is acquired by the judgement processing.
