A Novel Picture Coding Using Colorization Technique

Colorization is a computerized process of adding color to a black and white print, movie and TV program. The authors have proposed automatic colorization algorithms by giving a partial color to a monochrome image. This paper focuses on the colorization process which can produce a color image from a monochrome image with a small number of color pixels, and proposes a novel color image coding algorithm based on the colorization technique. At first, luminance component is separated from an input color image. Next, selected color seeds from the original color image are sown on the luminance image domain automatically and the monochrome image is colorized. The sowing process is continued until the colorized image satisfies the desired quality. Finally, both orthogonal transform-coded luminance component and the set of color seeds are transmitted as coded data. The decoding can be performed by tracing the same colorization process. We confirmed that the colorization technique is effective to image coding through the experiments. Introduction With the recent spread of the internet and multimedia technologies, digital images play more and more important role in human visual communications. In order to reduce image data quantity, several image compression algorithms have been developed. Most of them, such as JPEG2000, make use of the spatial correlation. In natural images, a strong correlation can also be observed among tri-color signals. The authors have focused on luminancechrominance redundancies, and have been proposed novel images compression algorithms [1],[2]. This paper stands on the extension of those works, but uses a different approach. In this century, the study of “colorization” begins to attract attention. Colorization is a technology of coloring monochrome images automatically by giving a few hints of color. Welsh et al. proposed a semi-automatic algorithm by transferring color from a reference color image [3]. Levin et al. proposed an interactive colorization method by giving some color scribbles [4]. The authors have also proposed a few colorization algorithms by sowing color pixels [5]-[9]. This paper presents basic experimental results for applying a colorization technique to image data compression. Figure 1 shows the overview of the proposed coding scheme. At first, luminance component is separated from an input color image. Next, selected color seeds from the original color image are sown on the luminance image domain automatically and the monochrome image is colorized. The sowing process is continued until the colorized image satisfies the desired quality. Finally, both orthogonal transform-coded luminance component and the set of color seeds are transmitted as coded data. In the colorization process, our pixel-sowing colorization techniques [5]-[9] have an advantage than other techniques. Because it is difficult to extract color hint by using other colorization techniques such as reference images and color scribbles. Furthermore, the proposed technique meets the human visual system well in which the sensitivity of the eye to luminance detail is higher than that of chrominance detail. The detailed algorithm will be explained in the following sections. Figure 1. The proposed coding model using colorization technique. Colorization Algorithm In the proposed coding algorithm, all techniques in Refs.[6]-[9] can be used as the colorization process. In this paper, we briefly introduce a colorization technique in Ref.[8]. Let ) , ( y x I = be a pixel in an input monochrome image and let S be a set of color seeds. The color seeds, which are color pixels strictly, are given as a prior knowledge by the user. Therefore, position and coloring of those seeds are determined by the user. Note that the color must be chosen with keeping the luminance of the original monochrome pixels. Since we present our method in CIELAB color space, each monochrome pixel I is transformed into the luminance signal ) (I L . Note that other color space like YCC, YUV would work equally well. The color seed S is also transformed into a luminance component and a pair of chrominance components ) ( ), ( ), ( S b S a S L , respectively. In Ref.[8], a four-connected pixel I of the seed S is colorized by ) ( ), ( ), ( S b S a I L components in CIELAB color space, if the following partitioning condition satisfies: . ) ( ) ( Th S L I L < − (1) where Th means a threshold of the partition. The partition works for preventing error propagation. If the difference of luminance between adjacent pixels is larger than the threshold, the color propagation stops. Otherwise, the color propagation will be continued to the next adjacent pixels. When the propagation stops everywhere, a next color seed is sown on the remaining monochrome pixels. The sowing process will be continued until all pixels will be colorized. Figure 2 shows an example of colorization by using the algorithm in Ref.[8]. Figure 2(a) shows an input monochrome image with red circles which expresses the position of color seeds. Each seeds were sown at the center of the circle. In this example, seven seeds were finally sown on the monochrome image for colorizing all monochrome pixels. Figure 2(b) shows the colorized result. Image Coding Algorithm This section shows the proposed image coding algorithm, which consists of luminance component decomposition, color seed selection, color propagation and data coding. Decomposition of the Luminance Component First of all, luminance component is separated from an input color image. In this paper, we would explain the colorization process in CIELAB color space, the color image transfers into L*a*b* planes here using the popular color transformation. Color Seeds Selection As explained in the previous section, in the colorization algorithm in Ref.[8], the position and color of seed pixels were given by the user. However, in the image coding process, it is required to select them automatically. Fortunately, the color of seeds can be determined by the input color image. Therefore, we should extract only the position of seeds automatically. Here we tested the following three methods for placing of seeds. A. Random Obviously, a random setting of seeds resulted in the worst and huge seeds were required to colorize all pixels, because it is independent of image color distribution and seeds were sown on isolated regions. It is required to propagate many pixels by each color seed for reducing data size. B. Selection from high luminance histogram This method assumes that pixels in the same region have almost the same luminance. According to the assumption, a pixel with the luminance of high histogram may belong to a large region on the image. Seeds will be selected depending on the present luminance histogram. C. Box center at higher pixel density in CIELAB space To select more reliable seeds depending on the input image, we generated M=m pieces of rectangular boxes in CIELAB color space surrounded by the regular lattice points inside the min-max color ranges of image color distribution. The image color distribution is partitioned by a unit box with the size of ∆a × ∆b × ∆L m b b b m a a a m L L L