To provide an image processor which improves reproducibility in a low density part or high density part when binarizing image information by means of an error diffusion method and does not require a complicated processing circuit.

This image processor converts multi-gradation data to two- gradation data by using the error diffusion method. A pixel position information converting means 2 converts a pixel position information of a pixel of interest to another pixel position information. A first noise generating means 3 generates several kinds of noise by using the information converted by the pixel position information converting means 2. Besides, a second noise generating means 4 generates several kinds of noise by using both the position information and the pixel density of the pixel of interest. Then, a threshold processing means 1 for adding a correction value based on noise outputted by respective noise holding means to a threshold held by a threshold fluctuating means 5 performs binarizing processing with the threshold added with the correction value based on the noise. An error distributing means 6 distributes errors which are generated at binarization, to nearby pixels by weighting.