An image processing apparatus, which stochastically binary-converts lower-bit data of a pixel of interest to improve image quality in a highlight portion, as well as in a low-density portion, is provided. In an image processing apparatus for quantizing multi-value image data to binary data, multi-value image data of a pixel of interest is divided into upper- and lower-bit data by a calculator (24). The lower-bit data is binary-converted by a comparator (25) using a normalized uniform random number generated by a random number generator (26) as a threshold value. The binary-converted data is added to the upper-bit data of the image signal of the pixel of interest by an adder (22), and the sum data is binary-converted by a comparator (18) using one of a plurality of weighted average values of a plurality of pixels near the pixel of interest, which values are calculated by a calculator (600), on the basis of binary data of pixels, which have already been binary-converted before the pixel of interest. At this time, the density is preserved by distributing a density error generated upon binary conversion to non-binary-converted pixels near the pixel of interest by a circuit including a subtracter (17), an adder (22), and the like.