Image fusion of smart phone equipped with high- and low-resolution dual cameras

• Main Authors: Yu-Hui Lin, 林育暉
• Other Authors: John Y. Chiang
• Format: Others
• Language: zh-TW
• Published: 2016
• Online Access: http://ndltd.ncl.edu.tw/handle/8f4868

碩士 === 國立中山大學 === 資訊工程學系研究所 === 104 === The resolution of digital imaging increases from 300K to 1.8M pixels nowadays. Limited by the thin and light-weight casing of smart phone, only small image sensors can be mounted, causing fairly restricted amount of light collected on the image sensing area. With the increase in resolution, even though more image detail can be acquired, yet accompanied with the amplification of noise, decrease in total incident light and sensitivity due to the small pixel sensing area. All these lead to pitfalls of image darkness, blurring, low contrast, etc. Attempts in increasing the exposure time or ISO inadvertently causing motion blur or large amount of noise, severely degrading the image quality. In light of the aforementioned limitation of the single-lens setup, the smart phone equipped with dual-lens is a promising alternative. Through two lens with different spatial resolution, aperture or focal length, the lens can be operated independently with different sets of parameters to acquire two images taken at the same time, each with the corresponding prominent features. These prominent features can be adaptively integrated through image fusion technique to form one image with better quality. This project will employ multi-resolution technique to fuse two images captured through low- and high-resolution dual lens to render a more consistent image in different lighting environments. By integrating the low noise level and low motion blur present in low resolution image, and good image detail in the high resolution counterpart, image fusion can preserve the prominent features present in two images to compensate the inherent weakness present in each type of lens. The image detail acquired through high-resolution lens, and the luminance information captured by the low-resolution one can be combined to produce a more desirable composite image, even in a low-lighting scene. The images acquired will be transformed to frequency domain by multiscale transformation. Based on the frequency bands, different fusion rules will be applied to adaptively combine the frequency components. The low frequency regions will be evaluated to estimate the overall luminance to determine under- or over-exposure, texture, noise and just noticeable difference (JND). The corresponding fusion weighting will be derived correspondingly. Region maximum will be preserved in the high frequency regions to preserve subtle variations. Based on the fusion rule and weighting, the features prominent in each frequency band will be combined, and then inverse transformed back to spatial domain. Through the integration of redundant information, image accuracy, consistency and quality can be substantially improved. The drawbacks inherent in each type of lens can be compensated by adopting the prominent features in the counterpart to maintain image detail and light sensitivity, decrease motion blur and noise level, especially in low lighting environment, while at the same time enhance detail and contrast of dark regions within an image. Keywords: dual lens, image fusion, fusion rule, multiscale technique