Feature-preserving multilevel halftoning based on threshold decomposition

Abstract

Digital halftoning is a technique that converts a continuous image into a bilevel image and is widely used in printing applications. Nowadays, printing technology has advanced a lot and printers are able to print images with ink dots of different intensity levels. Accordingly, a technique for converting a continuous image to a multi-level image is needed. Such a process is generally referred to as multilevel halftoning or multitoning. Theoretically, multilevel halftoning can be implemented with conventional binary halftoning techniques after replacing their bi-level quantizers with a multilevel quantizer. However, this straightforward extension does not work properly and generally introduces banding artifacts in their outputs. Threshold Decomposition is a widely used technique to remove banding artifacts. It decomposes a continuous image into energy layers, converts them to binary halftones with a conventional binary halftoning algorithm one by one under a stacking constraint, and combines the resultant binary halftones to form a multilevel output. Though this approach provides flexibility and is able to remove banding artifacts effectively, the operation flow introduces a bias to favor the black dots and makes it difficult to preserve both the dark and bright spatial image features in the resultant multilevel halftones. This work addresses this bias issue and proposes several solutions to remove the bias. Two of them are dedicated for producing three-level multilevel halftones in which the input continuous image is decomposed into two energy layers. In the first solution we process both energy layers simultaneously in an interleaving manner. In the second solution we combine the two energy layers to form a complex energy plane for being processed such that we can take both layers into account at any time of processing. At the end, the idea of the first solution is generalized to take care of more than two energy layers without introducing the bias such that one can produce multitones of any number of output levels. Simulation results show that the proposed solutions can eliminate the bias effectively and produce multitoning outputs of better quality as compared with the conventional multitoning algorithms by preserving the spatial image features faithfully.