From halftoning to compressing halftone images

Abstract

In this work, we address two issues in bi-level image processing including digital halftoning and lossy compression of halftones. In particular, two digital halftoning algorithms based on multiscale error diffusion and one lossy compression technique based on vector quantization are proposed. Multiscale error diffusion is an advanced halftoning technique proposed recently. This technique can solve the problem of directional hysteresis, which is a common problem of conventional error diffusion algorithms. However, this technique is implemented in a frame-oriented approach and hence the computation and memory requirement is huge. This makes it impossible for real-time applications. A new digital halftoning algorithm based on multiscale error diffusion is proposed in this work to solve this problem. This algorithm is row-oriented and hence can reduce an amount of computation effort and processing latency as compared with conventional schemes. In conventional multiscale error diffusion algorithms, pixels are handled one by one according to a rule of maximum intensity guidance. The idea behind this guidance is that the pixel of the highest intensity should be taken care of first as they are more important. This guidance introduces a bias towards the assignment of white dots. Besides, we found that the positions of minority dots in a local region are more critical in preserving local image features and hence they should be handled first. In this work, we proposed an algorithm based on these observations. Without any preprocessing step, the proposed algorithm is able to preserve local feature details, which is critical to the output quality especially when the resolution of the output is limited by the physical constraints of the display unit. Besides, a simple yet elegant arrangement is also proposed in the algorithm to effectively remove the boundary effect that appeared in most conventional multiscale error diffusion algorithms. Conventional coding schemes for halftone compression are lossless and their achievable compression ratio is low. The lossy compression technique introduced in JPIG2 standard transforms a bi-level halftone image into a gray level one and then downsamples it to reduce the data size. The major data reduction is achieved by downsampling, which results in a reduction of spatial resolution. To preserve the spatial resolution, a new lossy compression technique based on vector quantization is proposed in this work. In this compression technique, the input image is partitioned into a number of nonoverlapped blocks. For each block, the encoder selects a codeword with minimum distortion subject to a criterion based on a human visual system model and then diffuses the error to its neighbor blocks. As a result, the overall error within a small region can be cancelled out. Simulation results show that the proposed method is able to achieve a compression ratio around 4 to 5 and its coding performance is both objectively and subjectively better than another algorithm which was claimed to be better than JPIG2.