Development of a New Object Recoloring Technique

Abstract

The processing on noisy images is quite a difficult task, whether they are to be segmented or recolored. If the noisy images are lit by multiple illuminants, the issue of processing such images is more tough. It also results in decreased visibility and appearance of the images resulting in hiding the significant details of the image. For this purpose, a simple and effective image recoloring approach is proposed for smooth images and images that have been corrupted with gaussian noise. Moreover, these images are lit by two illuminants. The proposed technique utilizes the concept of intrinsic image decomposition. Decomposing an image like this has taken a spike in the existing years. It applications include object segmentation, object recognition, color constancy algorithms and various other tasks. In image editing, recoloring plays an important role. There is a very simple work done on recoloring in which objects with only some white light on them are recolored and other illuminants are ignored. Multiple illuminants will be taken into account here. The image is processed in three parts in the proposed technique. First of all, image is divided into two parts i.e. structure and texture. Then, recoloring is done using modified multi illuminant dichromatic model on the structure part. This recolored image is then combined with texture of original image. Lastly, noise removal methods such as bilateral filter and others are applied on different part of the image. Modified sharpening and contrast adjustment technique is applied at the end of the image for visually pleasing results. Experiments done on images have yielded accurate results using this framework. Better recoloring and plausible illuminant estimation has been achieved than previous approaches. The image shows results in desired and highly effective way.