Sparse identification of contrast gain controlin the fruit fly photoreceptor and amacrinecell layer
What is the science
Have you ever wondered why when you use your camera to take a picture of the outside from within a dark room, either you can either see the outside light area very easily, or you can see the darker room easily. But not both. However you personally can see both the outside and the inside perfectly clear? This is called contrast gain control. Your visual system is tuned perfectly to handle these extreme changes in contrast in a given image. But it is not just humans that have this ability. Flies also have this ability. In the paper Sparse identification of contrast gain control in the fruit fly photoreceptor and amacrine cell layer, Dr. Aural Lazar, Dr. Nikul Ukani, and Dr. Yiyin Zhou, have built a model of a fly visual system that is especially good at detection and correction of low light contrast in an image.
What did they do
Here they are going to use a complicated piece of mathematics called the Volterra series. A Volterra series is to convolution as a Taylor series is a monomial function. A Taylor series is just a polynomial. Or a sum of different powers of a monomial. However, a Volterra series is just a sum of repeated convolutions. These repeated convolutions are akin to the polynomials in the Taylor series. What this allows the authors to do is perform divisive normalization with second order Volterra series. This allows them to estimate incredibly nonlinear functions, and create a very powerful algorithm to upregulate contrast in lowlight conditions and down regulate contrast in bright lights.
What did they find
Their algorithm did very well. When a camera takes a picture in a low light environment and the resulting image appears black. But buried in that image is the actual information. Simply using the logarithm of the pixel value, shows us that the picture is a tree, but the resulting contrast is still not good. However when the Volterra series and divisive normalization is applied the result is crystal clear. It is a very impressive algorithm, and well deserving of the complex math that it uses.
What's the impact
Contrast control is very important in robotic applications. Even when a robot passes under a tree or a cloud the resulting shadow can mess with its ability to see. Plus in some rescue situations there can be dark hallways that have areas well illuminated by flames, or outside light. Therefore having an algorithm that can adjust contrast, across many different light levels is a key area of improvement in computer vision. Algorithms like this are a key step in this direction.
Have you ever wondered why when you use your camera to take a picture of the outside from within a dark room, either you can either see the outside light area very easily, or you can see the darker room easily. But not both. However you personally can see both the outside and the inside perfectly clear? This is called contrast gain control. Your visual system is tuned perfectly to handle these extreme changes in contrast in a given image. But it is not just humans that have this ability. Flies also have this ability. In the paper Sparse identification of contrast gain control in the fruit fly photoreceptor and amacrine cell layer, Dr. Aural Lazar, Dr. Nikul Ukani, and Dr. Yiyin Zhou, have built a model of a fly visual system that is especially good at detection and correction of low light contrast in an image.
What did they do
Here they are going to use a complicated piece of mathematics called the Volterra series. A Volterra series is to convolution as a Taylor series is a monomial function. A Taylor series is just a polynomial. Or a sum of different powers of a monomial. However, a Volterra series is just a sum of repeated convolutions. These repeated convolutions are akin to the polynomials in the Taylor series. What this allows the authors to do is perform divisive normalization with second order Volterra series. This allows them to estimate incredibly nonlinear functions, and create a very powerful algorithm to upregulate contrast in lowlight conditions and down regulate contrast in bright lights.
What did they find
Their algorithm did very well. When a camera takes a picture in a low light environment and the resulting image appears black. But buried in that image is the actual information. Simply using the logarithm of the pixel value, shows us that the picture is a tree, but the resulting contrast is still not good. However when the Volterra series and divisive normalization is applied the result is crystal clear. It is a very impressive algorithm, and well deserving of the complex math that it uses.
What's the impact
Contrast control is very important in robotic applications. Even when a robot passes under a tree or a cloud the resulting shadow can mess with its ability to see. Plus in some rescue situations there can be dark hallways that have areas well illuminated by flames, or outside light. Therefore having an algorithm that can adjust contrast, across many different light levels is a key area of improvement in computer vision. Algorithms like this are a key step in this direction.
Author: Alex White
Source: Aural Lazar, Nikul Ukani, and Yiyin Zhou. Sparse identification of contrast gain control
in the fruit fly photoreceptor and amacrine cell layer. Journal of Mathematical Neuroscience, 10:3(2020).
https://mathematical-neuroscience.springeropen.com/track/pdf/10.1186/s13408-020-0080-5
in the fruit fly photoreceptor and amacrine cell layer. Journal of Mathematical Neuroscience, 10:3(2020).
https://mathematical-neuroscience.springeropen.com/track/pdf/10.1186/s13408-020-0080-5
留言
張貼留言