Image conversion with lossy adaptive error diffusion

Abstract

Gray level pixel values in an image, each pixel value represented by c levels are quantized by applying a threshold level to each pixel value in the image to produce a pixel value having d levels, and applying a weighted portion of the value of the difference (the 'error') between the pixel value and the thresholded value to a predetermined set of neighboring pixels; for each neighboring pixel in the predetermined set to which the error term is to be applied, comparing the value of the neighboring pixel to each possible legal output value, and if any pixel value the predetermined set of neighboring pixels is equal to one of the legal output values, then the error term is not applied to that term, and is applied to remaining non-legal values. If all the neighboring pixels have legal values, a decision is made based on a look-ahead neighborhood (a set of pixels used to determine the fractional error allocation) as to whether the error term is discarded or preserved. In the most simple case, the look-ahead neighborhood is identical to the error distribution neighborhood (the set of pixels to which a fractional error can be distributed) and the error term is discarded when all of the neighboring pixels have legal output values. In a more general case, the look-ahead neighborhood extends over an area larger than the error neighborhood, and the error term is only discarded when additional criteria are met, e.g. when all pixels in the larger look-ahead neighborhood have legal values.