Local classification-based approach to generate halftone of scanned images with low computation and memory requirements

Cluster dot dithering is very useful halftoning for multifunction printers with electrophotographic printing process. However, there is an inevitable trade-off between spatial resolution and grey tone levels: sharpness and smoothness. In addition, it tends to produce undesired artifacts when printed images such as newspapers and magazines are copied. In this paper, we propose efficient, locally-content adaptive, and clustered dot halftoning, which improves image sharpness, increases text readability, suppresses artifacts, and maintain smoothness in scanned images from printed documents or natural images. We first split an image into 5x5 pixel blocks, and then each block is processed in raster scan order. Each pixel in the current processed block is classify into one of three categories by the process of non-smooth detection and halftone extraction. The disconnected characteristic of halftones is then used to separate halftone pixels from edge pixels. An adaptive approach to generate halftone images is used in each different type of block. Edge-enhanced cluster dot dithering is applied to edge blocks to reproduce sharp edges and also minimize block artifacts. The scaled and weighted factors are then used to determine the number and the position of black dots in the current processing block. To get the proper weighting and scaling values, we use average lightness and minimum square error from dithered images of several step inputs. Unlike edge block, preprocessing is performed before halftoning in halftone and complex blocks to recover the original continuous image from a halftone image. For halftone block, we estimate halftone resolution from the gradients at halftone pixels by three pairs one-dimensional derivative masks with different size, and then apply the average filter, which is chosen by the estimate of halftone resolution, to each pixel. For complex block, we group all pixels by virtual edge lines and then compute the average value of pixels in each group. The virtual edge lines are determined by edge direction and edge pixels along the edge direction, and these edge lines are used to separate a complex block into two or more groups which have different grey values in the original continuous-tone image. We use only one screen in an entire image to minimize block artifacts that appear when switching between different approaches at boundaries of different types of blocks. The proposed method is suitable for hardware implementation because it requires a small amount of memory and simple operations. Our experiments show that text readability and edge sharpness are enhanced while image smoothness are reproduced