基于参考纹理与自身色彩的图像修复

Image inpainting using reference image texture and distress image color

传统的图像修复工作仅仅利用破损图像本身的信息完成,破损面积较大并且结构比较复杂时,破损图像不能提供足够的信息导致修复效果不理想。针对这个问题提出了基于参考图像纹理与破损图像自身颜色的修复算法。该算法在图像库中通过图像检索智能筛选相似参考图像,并选择最优区域填充破损图像区域,利用参考图像样块与自身未破损区域的纹理信息保证修复边界的平滑性,再结合颜色迁移与扩展算法使破损图像修复区域与完好区域的色彩协调一致。实验结果表明新提出的修复算法使得图像修复区域过渡更加自然,能在视觉上有较好的效果。

While dealing with a target image of large damaged area and complex structure, the traditional image restoration algorithms cannot get sufficient information from it, leading to unsatisfactory repair effects. An algorithm based on the texture of reference image and the color of target image was proposed to overcome this defect. First, the suitable reference image was obtained by using intelligent search from similar images in image database and the template fill was performed on the damaged regions by selecting appropriate content from reference image. Then, the restored boundary was smoothed by the texture information of both reference image and target image. Color transfer and colorization was implemented to make the appearance of repaired part in accordance with its surroundings. The experimental results show that this approach can obtain better results and keep smooth region boundaries in visual effect.