Input pixels are input to a thinning processing circuit. When thinning is performed at the thinning processing circuit, minimum values are selected in order to reduce the pixel data, whereby loss of thin lines is prevented. The thinned image data are input to a binarization processing circuit. Although the image having been thinned through selection of minimum values becomes dark as a whole, collapse of characters can be suppressed, because the binarization processing circuit dynamically changes a binarization threshold by employing, as the binarization threshold, the average of maximum and minimum values in each local region. In a density flat portion, the local maximum value and the local minimum value assume the same value. Therefore, when the difference between the local maximum value and the local minimum value is not greater than a predetermined range, binarization is performed by use of a fixed threshold value. Instead of local maximum and minimum values, a local average value may be used as a binarization threshold value.