Back to results list
Show full item record
Please use this identifier to cite or link to this item:
|Title:||Feature-preserving processing techniques for color filter array images||Authors:||Chung, King-hong||Degree:||Ph.D.||Issue Date:||2009||Abstract:||To reduce cost and size, most digital cameras capture scene with a single-sensor image acquisition system in which the sensor is coated with a Bayer color filter array (CFA) and samples only one of the three primary colors (red, green and blue) at each pixel. The mosaic sensor raw data, referred to as Bayer CFA image, is then converted to full-color image by estimating the two missing components of the pixels. This process is called color interpolation or color demosaicing, and it is critical to a digital camera as it determines the output quality of the camera. The demosaiced full-color image may then be enhanced with a post-processing step to attain a visually pleasing output before being compressed for storage. Recently, such a demosaicing-then-compression arrangement is found to be sub-optimal from the compression point of view because eventually the compression process has to remove the redundancy introduced in the demosaicing process. To solve this problem, the system is modified by adding an additional processing branch or by replacing the original pipeline with a new one to allow direct compression of the Bayer CFA image before demosaicing. As a result, more sophisticated demosaicing and post-processing algorithms can be applied and carried out offline in a powerful computer to attain a higher quality output. This thesis investigates various processing algorithms to improve the performance of an advanced pipeline in terms of quality and complexity. It deals exclusively with raw Bayer CFA images and focuses on addressing the problem of color demosaicing, digital zoom and image compression respectively. Color demosaicing is one of the most important processes in a single-sensor image acquisition system as it turns a CFA image into a full-color image. In this thesis, a feature preserving demosaicing algorithm is first presented to restore a full-color image from a Bayer CFA image. This algorithm uses the variance of color differences as a supplementary criterion to determine the interpolation direction for estimating the missing green components. The missing red and blue components are then estimated based on the interpolated green plane. This algorithm can effectively preserve the details in texture regions and, at the same time, can significantly reduce the color artifacts. Simulation results show that the proposed algorithm produces superior demosaicing results both objectively and subjectively. Digital zoom is another common process performed in a digital camera. Basically it performs interpolation and hence employs similar signal processing concept as color demosaicing does. However, in a conventional system, the zooming process is generally carried out separately in the post-processing step. Accordingly, the information available on the raw sensor data is not always utilized consistently and efficiently to yield the enlarged output image. To remedy such inefficiency, a joint color demosaicing and digital zooming algorithm is proposed for digital cameras to produce a high quality zoomed output at a reduced computation cost. This algorithm directly extracts the edge information from raw sensor data for interpolation in both demosaicing and zooming to preserve edge features in its output. It allows the extracted information to be exploited consistently in both stages and also efficiently as no separate extraction process is required in different stages. This algorithm can produce a zoomed full-color image as well as a zoomed Bayer CFA image with outstanding performance as compared with conventional approaches which generally combine separate color demosaicing and digital zooming schemes in a straightforward manner. Simulation results show that our first proposed demosaicing algorithm can provide a high quality output. However, it may not be suitable for real-time realization in a low-profile camera as its computational complexity is comparatively high. Inspired by the idea of sharing edge information in the realization of the joint demosaicing and zooming algorithm, a more advanced and efficient demosaicing algorithm is proposed for constructing a full-color image. This algorithm exploits a new edge-sensing measure called integrated gradient (IG) to effectively extract gradient information in both color intensity and color difference domains simultaneously. This measure is reliable and supports full resolution, which allows one to interpolate the missing samples along an appropriate direction and hence directly improves the demosaicing performance. By sharing IG in different demosaicing stages to guide the interpolation of various color planes, it guarantees the consistency of the interpolation direction in different color channels and saves the effort required to repeatedly extract gradient information from intermediate interpolation results at different stages. An IG-based green plane enhancement is also proposed to further improve the efficiency of the algorithm. Simulation results confirmed that this proposed demosaicing algorithm outperforms other up-to-date demosaicing algorithms including our first proposed demosaicing algorithm in terms of output quality at a complexity of around 80 arithmetic operations per pixel. The lossless compression of Bayer CFA images is finally addressed in this thesis. Since a Bayer CFA image acts as a "digital negative" and can be used as an ideal original archive format, lossless compression of Bayer CFA images is highly preferred especially in the field of high-end digital photography. However, as different color channels are interlaced in a Bayer CFA image, the spatial correlation of a Bayer CFA image is seriously damaged. Conventional lossless image compression schemes such as JPEG-LS and JPEG2000 cannot make use of the spatial correlation to attain a good compression performance. A prediction-based lossless compression scheme is proposed in this thesis to effectively de-correlate the pixel redundancy in a Bayer CFA image. This scheme exploits a context matching technique to rank the neighboring pixels when predicting a pixel, an adaptive color difference estimation scheme to remove the color spectral redundancy when handling red and blue samples, and an adaptive codeword generation technique to adjust the divisor of Rice code when encoding the prediction residues. Simulation results show that the proposed compression scheme can achieve a better compression performance than conventional lossless Bayer CFA image coding schemes.||Subjects:||Hong Kong Polytechnic University -- Dissertations.
Image processing -- Digital techniques.
Color photography -- Processing.
|Pages:||xviii, 150 leaves : ill. (some col.) ; 30 cm.|
|Appears in Collections:||Thesis|
View full-text via https://theses.lib.polyu.edu.hk/handle/200/4355
Citations as of Jun 4, 2023
Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.