Interregional Adaptive Image Color Transfer Algorithms And Performance Evaluation Based On

Color transfer is an important technique of image non-photorealistic rendering and image editing, it not only can realize color transfer between color images but also can colorize grayscale images. It can be applied in computer animation, video editing and image stylized rendering. Color transfer methods include global methods and local adaptive methods. Local adaptive approaches can match local features between images better, and thus they get wider attention. However, if you want to get better local adaptive color transfer results, a lot of manual intervention are required which increase the burden on the user and limit the practical application of these color transfer algorithms. Therefore, research on color transfer algorithm based on less interactive information and regional adaptive color transfer algorithm are particularly important. In this paper, We mainly research regional adaptive color transfer algorithms, it mainly includes four aspects:apply interactive segmentation and match by marking to realize regional adaptive color transfer; apply super pixel and fast cascade feature matching scheme to realize adaptive and automatic colorization of grayscale image; apply multiple clues features predictions to realize adaptive colorization of grayscale image and the quantificational evaluation on the performance of color transfer algorithm using mathematical methods.The main work of this paper includes:(1) Through the study of a variety of color transfer algorithms and traditional image quality evaluation methods, this paper propose a metric of quantificational evaluation on the performance of color transfer algorithm using mathematical methods which is suitable for color transfer algorithm, this method is Perception Color Naturalness Ration(PCNR). It uses a mathematical formula to calculate a number that can represent the quality of color image generated by the color transfer algorithm. Thus, it can overcome the drawback that traditional subjective evaluation methods are not objective enough, and makes the evaluation result consistent with human visual perception better.(2) To overcome the drawbacks of color misrepresentation in classical global color transfer approaches, a local adaptive color transfer algorithm based on interactive segmentation and match by marking is designed to deal with the complex nature images. The pipeline of the algorithm contains three phases. In the first phase, the interactive segmentation is applied to the reference image and target image. In the second phase, the regions with similar scene of the two segmented images are grouped into matched region pairs by marking scribbles. Finally, local adaptive color transfer is carried out between the matched region pairs. Experimental results demonstrate that the proposed method can reduce color misrepresentation in the color transfer which improves the naturalness and the local consistency of the result image.(3) To overcome the limitations of the color transfer approaches based on user interaction in the practical application, we deeply research the adaptive and automatic colorization approach based on super pixel features and fast cascade feature matching shceme, and improve it on the basis of it. The improvement of it is we use SLIC approach to generate super pixels. The pipeline of this algorithm contains four phases. In the first phase, SLIC approach is applied to the reference image and target image to generate their own super pixels, then extract brightness, standard deviation、Gabor、SURF features for each super pixel. In the second phase, apply fast cascade matching scheme to find matched super pixel pairs. In the third phase, color transfer is carried out between the matched super pixel pairs and then extends to the entire grayscale image. Finally, voting in the color space of the middle image is used to identify the wrong color assignments, and then perform reassignment. Experimental results demonstrate that the improved approach realizes a fully automated grayscale image colorization, and it is faster, it also can achieve higher feature matching and color naturalness of grayscale images,the effect is better, and needs less time.(4)Study of adaptive and automatic colorization grayscale image algorithms using multiple features predictions. The pipeline of this algorithm contains four phases. In the first phase. Discrete Cosine Transform approach is applied to the reference image and target image to extract features for each pixel, then dimensionality reduction is carried out using PCA. In the second phase, we estimate the probability distribution of all possible colors for each pixel to be colored. In the third phase, an energy function is constructed based on the probabilities, and then get final colored image through optimization using adaptive algorithm graph cut.