Restore blurred and defocused images using a Wiener filter. Implementation on C ++ OpenCV

In the continuation of the article about the restoration of defocused and blurred images I want to share my results with the Wiener filter for restoring real images. OpenCV 3.4 was used as an image processing library. Camera - Nikon D320, Nikon DX AF-S NIKKOR 18-105mm lens, defocusing was done manually, shooting was done without a tripod.


The results of recovering defocused images


During defocusing, the distorting system is well approximated by a cylindrical point spread function (PSF) of radius r.


o_psf
Cylindrical PSF


Below are the results of the restoration of three real defocused images of the same object (book page). The survey was carried out without a tripod from a distance of approximately 50 cm. The degree of defocusing of the lens manually increased from frame to frame. The Wiener filter parameters r and the signal-to-noise ratio (SNR) were selected manually in such a way as to ensure the best visual quality of the recovery. To compensate for edge effects, a gradual decrease in the brightness of the image at the edges is made.


o1f
Picture A


o1r_gamma-var_640
Result of image restoration A. r = 53, SNR = 5200


o2f
Picture B


o2r_gamma-var_640
Result of image restoration B. r = 66, SNR = 4400


o3f
Image C


o3r_gamma-var_640
Result of image restoration C. r = 102, SNR = 7100


It is seen that even with a significant defocusing, the readability of the text is almost
completely restored.


Results of restoring blurred license plate images


Image blur occurs when the camera and the object move relative to each other during exposure. Consider only the case when the subject is moving linearly relative to a fixed camera. In this case, the distorting system is well approximated by the PSF in the form of a segment, which is directed along the motion of the object. Such PSF is defined by two parameters: L length and THETA blur angle.


m_psf
PRT with linear lubrication


Below is a distorted image of two cars, obtained with insufficiently short exposure, which led to a noticeable blurring.


m1
Distorted image of two cars


Below are the results of the restoration of the numbers of both cars using the Wiener filter. The values ​​of the parameters L, THETA and SNR were selected in such a way as to ensure the best visual quality of the car number recovery.


m2
The result of the restoration of the light car number. L = 78, THETA = 15, SNR = 300


m3
The result of the restoration of the dark car number. L = 125, THETA = 0, SNR = 700


It can be seen that even with significant lubrication it is possible to restore the readability of
car numbers .


The filtering algorithm is implemented in C ++ OpenCV as a console application.
Source codes can be found at the links below.


https://github.com/VladKarpushin/out_of_focus_deblur/tree/full_color_version
https://github.com/VladKarpushin/motion_deblur
https://docs.opencv.org/4.0.0-alpha/de/d3c/tutorial_out_of_focus_deblur_filter.html
https : //docs.opencv.org/4.0.0-alpha/d1/dfd/tutorial_motion_deblur_filter.html


Literature


  1. RC Gonzalez, RE Woods. Digital image fundamentals. 1987
  2. I.S. Gruzman, V.S. Kirichuk, V.P. Kosykh, G.I. Peretyagin, A.A. Spector. Digital image processing in information systems. 2000
Tags:

Also popular now: