【摘要】 针对Criminisi图像修复算法的修复次序没有充分利用图像结构特征等特点,提出一种使用HSI颜色空间改进的基于样例的图像修复算法。通过在计算优先权值函数时将图像映射到HSI色彩空间,引入色调梯度和饱和度梯度等概念,并在信息量参数中引入梯度的均值和方差等统计量,充分提取图像的结构特征,使得修复次序趋于合理。另外,通过利用图像的邻域相似特性,减少了修复过程的运算量。对比实验结果表明,该算法对各类图像的修复质量都有较好的改善。更多还原

【Abstract】 Aiming at the problem of Criminisi’s image inpainting algorithm that the inpainting order does not take full advantage of the structure properties,we propose an exemplar-based image inpainting algorithm which uses HSI colour space to modify. By mapping the image onto HSI colour space when calculating the priority weighting function,the algorithm introduces the concepts of hue gradient and saturability gradient,as well as introduces to information quantity parameter the statistics of mean and variance of gradient,thus fully extracts the structure properties of image and makes the inpainting order tend to be rationalised. Besides,by utilising the neighbourhood similarity of image,the algorithm reduces the calculation amount of the inpainting process. Results of contrast experiment show that this modified algorithm achieves better improvement in inpainting qualities on various kinds of images.更多还原