Abstract: Image seamless cloning technique is an important filed in image editing region. In previous work, researchers separated the target object from source image and then embed it into the target image without consideration of whether the source image and the target one arc semantic matching. A novel and complete image cloning system was presented,in which Gist descriptors were used to describe image's scene semantics and semantic matching technique was applied to arbitrary cloning method to give it a realistic significance, which is called synonymous image cloning system. In consideration of the large time complexity of the semantic matching process, CUDA parallel programming model and OpenMP multithread programing method of the heterogeneous multicore environment were used to accelerate the process and effectively improve the performance of overall system. Experiments show that this system guarantees higher user involvement,stronger practicability and ensures the users' satisfaction of the cloning speed and effect.

Key words: Image editing, Scene recognition, Semantic matching, Arbitrary cloning