Image blending using graph cut method for image mosaicing

Abstract

In this research work, feature based image mosaicing technique and image blending using graph cut method has been proposed. The image mosaicing algorithms can be divided into two broad categories. The direct method and the feature based method. The first is the direct method or the intensity based method and the second one is based on image features. The direct methods need an ambient initialization whereas, Feature based methods does not require initialization during registration. The feature based techniques are followed by the four primary steps: feature extraction, feature matching, transformation model estimation, image resampling and transformation, and image blending. Harris corner detection, SIFT and SURF are such algorithms which are based on the feature detection for the accomplishment of image mosaicing, but the algorithms has their own limitations as well as advantages according to the applications concerned. The proposed method employs the Harris corner detection algorithm for corner detection. The features are detected and the feature descriptors are formed around the corners. The feature descriptors from one image are matched with other image for the best closeness and only those features are kept, rest are discarded. The transformation model is estimated from the features and the image is warped correspondingly. After the image is warped on a common mosaic plane, the last step is to remove the intensity seam. Graph cut method with minimum cut/ maximum flow algorithm is used for the purpose of image blending. A new method for the optimisation of the cut in the graph cut has been proposed in the research paper.