서강대학교

초록/요약

Multi-image fusion is a process of combining images that have different characteristics each other. This dissertation focuses on high dynamic range imaging (HDRI) and exposure fusion method using multi-exposure images, that is, low dynamic range (LDR) images with different exposures. HDRI and exposure fusion method are typical multi-image fusion methods to enhance the dynamic range (DR) of an image. Also, the artifacts such as noise, blur, and false color are reduced by multi-image fusion. In this dissertation, the proposed Tone mapping (TM) method uses the color correction function (CCF) and image decomposition in HDRI. The CCF in the proposed TM is derived from the luminance compression function with the color constraint under which the color ratios, between the three color channels of the radiance map and the DR compression term, are preserved and color saturation is controlled. The proposed CCF is developed to locally perform luminance compression and color saturation control in local TM. For image decomposition, a bilateral filter is used and adaptive weight is applied to the base layer of the luminance. Also, this dissertation proposes a local TM algorithm, in which the high dynamic range (HDR) input is segmented using the K-means algorithm and a display gamma parameter is set automatically for each segmented region. According to the bilateral filtered luminance, an image is divided into a number of regions using the K-means algorithm. The display gamma value is set automatically according to the mean value of each region. Then, the tone of HDR image is reproduced by a local TM method with adaptive gamma value. Sometimes high sensitivity setting is needed for capturing images in low light condition such as a dim indoor and night scene. However, current digital cameras do not produce a high-quality HDR image when noise occurs in low light condition or high sensitivity setting. In this dissertation, we use a noise reduction method in generating radiance map using a set of LDR images with different exposures. And we propose a TM algorithm by considering the remaining noise in the radiance map. First, the radiance map is decomposed using the wavelet transform, and a bilateral filter and soft-thresholding is used in a multiresolution framework. The coarse-grain noise in high international organization for standardization (ISO) image becomes fine-grain as the image is decomposed further into its subband. It is difficult to reduce the coarse-grain noise at the highest level, however it is possible to reduce the fine-grain noise at a lower level in a multiresolution framework. And we render the tone-mapped color image using the proposed color saturation control parameter, which is set automatically according to the compressed luminance value. Finally, this dissertation proposes an exposure fusion method of degraded LDR images, which are degraded by sensor noise, blur, false color, and etc. We extract the reliable pixels and regions from degraded input LDR images with short- and long-exposure time. For reliable pixels and regions extraction, input LDR images are decomposed using the wavelet transform. In each multi-level and multi-resolution, the detail components in the short-exposed image and the color components in the long-exposed image are fused. Therefore, the artifacts such as noise, blur, and false color are reduced and DR is enhanced in the fused image. Computer simulations with various sets of real LDR images show the effectiveness of the proposed TM algorithms and exposure fusion algorithm in terms of the visual quality as well the local contrast. The proposed algorithms enhance the contrast, detail, and color, and reduce the noise, blur, and false color of the fused image. It can be used for contrast and color enhancement in various display and acquisition devices. Further research will focus on the development of the real-time algorithm and the development of the video enhancement algorithm.