Image Restoration-Brief Description

2. Method

3. Others

refer to

Image inpainting is a technology that uses the available spatial information of adjacent areas to fill the missing or damaged areas in an imperceptible way, and it is a research topic of image processing. Image inpainting is an important research branch of image processing. Image inpainting is an ancient method to inpaint images, which are damaged and discarded due to scratches. It plays an important role in computer graphics, preserving historical heritage and eliminating unnecessary objects. Several image inpainting algorithms have been proposed, such as partial differential equation-based inpainting algorithm, sample-based inpainting algorithm, bilateral filtering inpainting algorithm and fast digital inpainting algorithm. By propagating the known information in the isochromatic direction, the lost information can be reconstructed. Isoillumination lines are lines with equal intensity. Image inpainting can be achieved by convolution with weighted average kernel.

Sometimes it is necessary to delete unwanted objects and scratches from the image. Unwanted objects can be logos, text, etc. Deleting can be done manually, but it is a tedious task. Unwanted objects and scratches cannot be erased directly, because erasing will leave white patches in this area. Fill in unnecessary and white patches by using adjacent spatial information. This method is called image inpainting.

With image inpainting, you can delete holes in the background as if the deleted object never existed. Therefore, the purpose of image inpainting is to fill the loopholes, create satisfactory information continuity, and make it difficult for neutral observers to identify the edited traces. Repairing can't reconstruct the original image, but it can fill in missing or unnecessary objects through parts very similar to the original image. Marcelo Bertalmio was the first person to put forward the concept of image restoration.

Image inpainting algorithms are widely used, including:?

A) Object removal: The user-specified object can be removed in a visually reasonable way by using recovery technology. ?

B) Scratch elimination: By applying an image restoration algorithm to the part containing scratches, the old image damaged by scratches can be restored.

C) Correction of damaged images during transmission: Images are often damaged in wireless transmission. By treating the missing part as a repair domain, the original image can be restored.

D) Produce amazing visual effects: In the image, amazing effects can be produced by restoration technology.

E) Text deletion: The patching algorithm can be used to eliminate unnecessary text on the image. ?

F) Other applications, such as red-eye correction, restoration of old damaged movies, compression, etc.

Restoration methods can be divided into diffusion-based restoration, sample-based restoration and convolution-based restoration.

Diffusion-based restoration method spreads the information of known images to unknown areas at pixel level. It uses the concepts of variational method and partial differential equation. This method can not provide satisfactory results for texture images, and at the same time, it will also produce blur. This method can be used if the area to be repaired is small.

In the sample-based restoration method, the missing region is filled by information from the surrounding known regions at the patch level. This method provides impressive results in restoring texture and repeating structure. However, their ability to reconstruct geometry without any examples is limited and not fully understood. This method overcomes the shortcomings of diffusion algorithm. The proposed image inpainting algorithm follows the following steps: 1) Calculate the priority of each block; 2) selecting the best patch; 3) Filling in sequence. Determine the method of filling holes according to the global image.

The convolution-based restoration algorithm [3] draws an image by convolving the neighborhood of damaged pixels with an appropriate kernel. They are very fast, but they are not effective at high contrast damaged edges. This method uses the gradient of the image to be repaired to calculate the mask coefficient. The algorithm is fast, iterative, easy to implement, and provides very sufficient results.

Oliveira et al. proposed a fast image inpainting algorithm using convolution operation. In their algorithm, the region to be repaired is repeatedly convolved with a predefined diffusion mask. This model is very similar to isotropic diffusion. In this method, the center weight of the diffusion mask is considered to be zero, because the relevant pixels in its original image are unknown. It can remove large objects in symmetrical background images without blurring, but it will fail or produce poor results when removing large objects in natural images.

Image inpainting is a technology to fill the missing data in the image, and it is also a research topic of image processing. It fills in damaged and lost data in an imperceptible way. The purpose of image inpainting is to reconstruct the missing area in a visually credible way and make it reasonable for human eyes. This paper discusses the repair methods of four papers, namely 1) repair based on global and local consistency; 2) restoration based on gated convolution; 3) Restoration based on edge learning; 4) Restoration based on foreground perception.

[ 1] ? Wang Xuewen, et al. Research progress of digital image inpainting technology [J]. IRJET, 2016,3 (1).

[2] ? Zhang Zhijun, et al. Overview of image inpainting technology. Texture synthesis, convolution and sample-based algorithm in image inpainting [J]. International Journal of Science and Technology Innovation Research, 20 15,1(7):100-106.

[3] ? Wang Xiaohua, Wang Xiaohua. Image Restoration Technology Based on Convolution [C]//2003 First International Communication Engineering Conference. 20 10: 130- 134.