How to deal with blurred picture

How to deal with blurred picture

2017200501005 格拉斯哥学院 甄雨欣

 

In our daily life , we are often bothered by some blurred pictures . Imagine that you are taking photos for a wonderful basketball match and you want to record the most excited moment of the game , but when you check the pictures , you find that what you have got are blurred one and you could not see details about the ball or the players . So this article will provide the basic way to deal with this problem .

The blurred image is usually caused by the relative movement of camera sensor , such as handshaking or fast motion of the object . This kind of problem could be solved by four steps , namely prediction , kernel estimation , deconvolution and final deconvolution .

Pictures are stored in computers , but they are not just performed in pictures but many digitals about the color , shape and many other information relating to the original image , which is actually matrix . More digitals are used to record the details of the image , more clear the image is . In order to deal with the blurred edge , a set of techniques would predict a strong edge and this process is actually called kernel estimation .

The photo would be ‘measured ’carefully in two aspect . Firstly , we would set image noise model . A linear function is used to represent the readout noise , which could be measured by comparing to an ideal noised free situation . The second data we need is relating to the motion , because there is a relative motion , so we just consider the camera motion . In addition , compared with the stable situation , the motion will be showed as a line rather than a point in the blurred picture and that line is actually the trace of the movement . Assuming that camera motion is dominated by rotation , the angular rotation at one point could be represented as a function . Although we get the characterizes of the blur , it is hard to just deal with the noise without affecting the truth scene . So the method we apply is doing a series of calculation to make sure we find the real blur caused by camera motion and we would just change this part .

What we have done before is to check the image , and now we would move to the changing processes .

In fact , we would apply more than one strategies depending on different situations and the first step is to determine the best strategy . Many factors will be considered and analysied to solve out a list of guidelines , which helps us determine the best capture strategy and reconstruction method for various environments . Align-and average(AA) and multi-image deconvolution(MD) are two approached in removing camera shake . However , AA is preferred to MD because many restrictions such as the effective rang of analog gain , the memory buffer in the camera system having a limited size and the capture time ,etc. In those environments , MD can provide an alter native with better performance . Finally , through suitable method , the blur could be removed  .

 

 

 

What’s more , such processes could speed up by doing the kernel estimation and latent image estimation individually to avoid using computationally inefficient priors for non-blind deconvolution . A method called gyro-based deconvolution could be applied to improve image quality in very low light environment . Also , similar process could solve the same problem in blurred videos .

 references

Fast Motion Deblurring

Gyro-Based Multi-Image Deconvolution for Removing Handshake Blur

Bundled Camera Paths for Video Stabilization