What is image degradation before image degradation/restoration modeling? The deterioration of image quality is called degradation. The forms of degradation include image blur, image interference and so on. Why is the image degraded? Images obtained by optical, photoelectric or electronic methods will be degraded to varying degrees; There are many forms of degradation. Such as sensor noise, unfocused camera, relative motion between object and camera equipment, random atmospheric turbulence, phase difference of optical system, scattering of imaging light source or light, etc.
Image restoration, like image enhancement, is to improve the visual effect of the image and facilitate the subsequent processing. Different from image enhancement, image enhancement method is more subjective, while image restoration is based on the reason of image distortion or degradation.
The key problem of image restoration processing is to establish degradation model. Without enough prior knowledge, we can use the existing knowledge and experience to establish and describe the mathematical model of the degradation process such as blur or noise, and restore the image according to the mathematical model of this degradation process.
Prior knowledge of image degradation process plays an important role in image restoration technology.
Generally speaking, there should be a prescribed objective standard for the quality of repair, so as to make some best estimate of the repair result.
In the field of signal processing, linear shift invariant system (or linear space invariant system) is often mentioned. Linear translation invariant systems have many important properties, and rational use of these properties will help us to deal with problems.
Find the filter transfer function, get the Fourier transform of the restored image through frequency domain image filtering, and then get the restored image through inverse transform.
Unconstrained recovery means that there are no other constraints except the minimization criterion function. Therefore, we only need to know the transfer function or impulse response function of the degraded system, and we can use the above-mentioned method to restore it. However, due to the ill-conditioned transfer function, the restoration can only be carried out in a limited area near the origin, which makes unconstrained image restoration have considerable limitations.
The reason of degradation is known: having prior knowledge of the degradation process, such as wanting to determine PSF and noise characteristics: that is, determining: and.
According to the physical process leading to ambiguity (prior knowledge):
In the process of digital image acquisition, due to the nonlinearity of the imaging system, the imaged image will be out of proportion or even distorted with the original scene image. This image degradation phenomenon is called geometric distortion.
Geometric distortion correction requires accurate geometric correction of distorted images. Usually, one image is used as a benchmark, and then another image is geometrically corrected.
Geometric distortion correction is generally completed in two steps: one is the transformation of image spatial coordinates-spatial transformation; The second is to re-determine the value of each pixel in the correction space-gray interpolation.
Spatial transformation: prevent image content from being fragmented (breaking straight lines)
Gray interpolation: the target image will need non-integer points of the original image.
After the geometric position correction of the image, the gray value of each image point in the correction space is equal to the gray value of the corresponding point of the corrected image. Generally, some pixels in the corrected image may be crowded together or scattered, and will not just fall on the coordinate points, so interpolation is often used to get the gray values of these pixels. There are two common methods.
1) nearest neighbor method:
The nearest neighbor method takes the gray value of the nearest neighbor among the four points adjacent to the pixel as the gray value of the point. As shown in the figure. The nearest neighbor method is simple in calculation, but its accuracy is not high, and the corrected image brightness has obvious discontinuity.
2) Bilinear interpolation method: