Named as one of CB Insights
100 Most Promising AI Startups 2023
In the world of digital imaging, three of the biggest barriers to clear, crisp imaging are low light, HDR, and high motion (fast-moving objects). But what exactly makes these three barriers so difficult to overcome? And what can AI-based software do to help?
Low light
Low light makes imaging difficult quite simply because, when creating an image, you need as much light as possible. Low light images tend to be of lower quality and with more noise because there are less photons to capture. The signal noise is created by the image signal processor (ISP) while processing the data into an image. One option to combat low light is to increase ‘gain’, which in turn increases brightness, but increasing gain will also increase the image noise.
HDR
HDR or WDR environments (High Dynamic Range or Wide Dynamic Range, respectively) present a unique challenge of a high range of very dark and very light areas within one image. While it’s meant to provide images with more of a variety of shades (that go beyond your standard “light” and “dark”) in order to make your images pop and look more realistic, it can often lead to blurriness and can reduce the quality of the overall picture.
The purpose of HDR/WDR is to balance out the brightness or contrast between lightness and darkness in a photo. For example, HDR/WDR may be used in a security camera placed at the entrance of a parking garage. This is because part of the image is constantly in the shade while the other is exposed to the sun.
Yet, while HDR/WDR is meant to offer a solution for these lighting issues, in reality, it continues to have its disadvantages. The difference between an original image and the image attained through HDR/WDR is just too strong. The balancing act used by HDR/WDR ends up hurting the quality of the image. In the end, you’re taking a risk regarding the final product and the loss of what made the original image unique.
High motion
Last but not least, high motion often results in blurred imaging because the objects being captured are moving so quickly. This problem is prevalent in lower resolution cameras and surveillance cameras, where sensors don't have enough time to catch and process the moving objects, thus resulting in a blur. Because there isn’t a proper balance when taking these types of photos, it’s almost impossible to get a clear, crisp image in these circumstances. As the shutter closes, the image captured is shifting and causes this blur. This issue can arise within multiple use cases that heavily rely on digital imaging, such as security, sports event safety, traffic monitoring, and self-driving cars. In all these areas, the need for high quality imaging of fast moving objects is extremely important and challenging to achieve.
These issues have plagued the imaging industry for years and as such, different solutions have been developed. However, these solutions are generally hardware-based, which results in the loss of valuable data caught by the sensors.
For example, as mentioned above, in order to compensate for low light, one solution is to increase gain which then increases brightness. However, adding gain results in more image noise. HDR/WDR imaging, which is meant to balance between the light and dark in images, often results in the loss of image quality or the over-emphasis of one area of the image.
Another direction for combating the barriers to crisp imaging is the adjustment of shutter speed. While increasing shutter speed may help with high motion images, it may darken a low light image even more. When dealing with different shutter speeds and their relevant uses, each problem requires a different solution.
For example, for a low light picture, a slower shutter speed is required, but a shutter speed that is too slow can ruin the image by capturing too much light. If a mistake with the shutter speed is made, you can end up with too much noise in the image and lower the overall quality. As a result, adjusting shutter speed is not an ideal solution. It’s just another example of a hardware-based temporary fix that does not maximize or optimize the data collected by the sensors.
These three imaging barriers are amplified by the fact that their “solutions” have one main thing in common: they are all hardware based and attempt to balance out the issue with external techniques.
The culmination of these barriers and their ineffective solutions has led to an increase in demand for AI capabilities that can completely revolutionize the way images are captured and processed.
By utilizing software image signal processing (software ISP) that leverages AI - as opposed to hardware ISPs - digital imaging can finally be done in such a way where sensor data is not damaged or lost.
With digital imaging at the forefront of how images are captured and its constantly expanding use cases, Visionary.ai has developed a unique process that overcomes these barriers.
Visionary.ai’s software ISP leverages all the raw data caught and collected by sensors, which can then be used to create crisp and clear images. The combination of AI and the migration of ISP from hardware to software creates a unique opportunity to take all of the data collected and use it effectively. Therefore, when capturing an image, nothing is lost and there is no need to compromise on quality in order to compensate for difficult imaging circumstances.