Development of Novel Techniques for Image Fusion with Improved Depth of Field and Dynamic Range

Abstract

Image fusion technique is a popular technique to extract scene information by combining complementary information from pre-registered captured images. Image fusion has many applications like remote sensing, medical imaging, surveillance and photography. Present work has proposed fusion techniques in the field of photography. In photography, captured images suffered from low depth of field (DoF) and lack of dynamic range. We have tried to address both the issues. The fusion can be performed in spatial domain or transform domain. In spatial domain direct operation is performed on images in order to extract complementary information whereas in transform domain, images are first converted to transformed images for processing and then converted back to synthesize fused image. Both techniques have their positives and negatives. Evaluation of fusion technique is also become important in order to evaluate the quality and to compare with stat-of-the-art techniques. Two type of evaluation techniques are available; subjective and objective evaluation. In subjective evaluation, group of trained peoples observe the image and rate them on given scale. Whereas in objective evaluation, mathematics based technique are used to test the quality. In subjective evaluation, different people have different views which sometime affect the rating of quality image. Therefore objective evaluation gives more concrete results. We have proposed three spatial domain fusion techniques by using filters. Three different spaces, block, feature and regions, are explored to increase depth of field and dynamic range. In block based technique, all pre-registered and captured images are splitted into local and global layers by using Neighbor distance filter. Local layer is processed with block based fusion whereas global layer is combined as weighted sum. Finally combined local and global layer are combined to synthesize fused image. It addressed both low depth of field as well as lack of dynamic range. In feature based technique, combination of recursive and Gaussian filter separates three features, constant, low varying and high varying features, from pre-registered and captured images. These features are processed separately for features with high activity level and further combined together to synthesize fused image. In region based technique, pre-registered and captured images are splitted into frequency and base layer by using recursive filter. The scene is divided into three regions based on intensity of captured images. The frequency layer is processed with region based fusion by using defined three regions, whereas base layer is combined as weighted sum. Finally fused image is synthesize by using processed frequency and base layer. Both feature and region based techniques addressed lack of dynamic range. The proposed techniques are tested on randomly selected standard data set. Objective evaluation based on probability, edges transfer are used to evaluate the quality of proposed techniques and compared with state-of-the-art techniques.