Image visibility is affected by the presence of haze, fog, smoke, aerosol, etc. Image dehazing using either single visible image or visible and near-infrared (NIR) image pair is often considered as a solution to improve the visual quality of such scenes. In this paper, we address this problem from a visible-NIR image fusion perspective, instead of the conventional haze imaging model. The proposed algorithm uses a Laplacian-Gaussian pyramid based multi-resolution fusion process, guided by weight maps generated using local entropy, local contrast and visibility as metrics that control the fusion result. The proposed algorithm is free from any human intervention, and produces results that outperform the existing image-dehazing algorithms both visually as well as quantitatively. The algorithm proves to be efficient not only for the outdoor scenes with or without haze, but also for the indoor scenes in improving scene visibility.

No abstract

Fusion of multi-exposure images for dynamic scenes often show ghost effect. This is mainly due to motion blur present in the image sequences or due to presence of totally new object. Detection and removal of such ghost effect plays an important role for automatically generating HDR images of dynamic scenes. In this paper, we present a simple, yet effective method for multi-exposure image fusion for dynamic scenes with new objects without ghost effect. The proposed method is carried out in four steps, viz.: weight map generation, new object detection, modification of weight maps and finally exposure fusion using modified weight maps. The experimental result show that our method produce HDR images without noticeable ghost effect.

No abstract

Image dehazing either using single visible image or using visible and near-infrared (NIR) image pair has seen growing interest in last decade for improving visibility in landscape photographs. In this paper, we propose a novel approach for image dehazing scheme using a pair of visible and NIR images. The dehazing mechanism estimates depth map and airlight color using the visible-NIR scene statistics and uses them to form a haze-free image. Experiments on a variety of hazy images demonstrate that our method achieves high degree of detail recovery over the existing image dehazing algorithms. The resultant images exhibit a very good blend of details, contrast and color. The proposed algorithm is less computationally demanding and is fully automatic. The results are superior in both visual as well as quantitative analysis compared to state-of-the-art image dehazing algorithms.