Image Processing Technology for Sharpening Images Captured by Network Cameras and Other Devices
A Technology to Raise Image Resolution
In ports and airports, advanced imaging systems that can quickly detect objects even in unfavorable shooting environments are required. Canon will contribute to solving social issues by utilizing image processing technologies that make use of the lens and sensor knowledge it has accumulated over the years.
October 16, 2023
Producing Clear Images with Image Processing Technology That Utilize Deep Learning
In recent years, advanced systems that are able to quickly detect the presence or absence of objects, or their suspicious movements, at any time of day or night, even at a distance, have been required at national borders, ports, and airports. Canon has been researching various image processing technologies that utilize deep learning, and is working to develop technologies that can transform images captured with network cameras into value-added images so that changes in objects can be clearly seen even in the dark or at a distance.
For example, if a network camera located outside captures a person walking in a dark environment at night, it will record a noisy, grainy image when the ISO is increased for shooting in the dark, and will leave a footage which is, as it is, difficult to see the person accurately. Therefore, Canon has developed an image processing technology that uses deep learning to predict noise and remove it.
Canon's image processing technology
What is needed for image processing using deep learning is a number of shooting cases in which the image you want to see is degraded by shooting under bad conditions. Canon has been developing and producing cameras for many years and is fully aware of the characteristics of lenses and sensors. Thus, Canon has a lot of knowledge about not only the cases of image degradation but also the causes including the fact that images taken in low light produce noise in part due to the structure of the sensor's electrical circuitry. Canon has accumulated numerous appropriate examples needed in the development of noise reduction method that process degraded images into "images you want to see." These examples enable objective image processing without relying on human experience or intuition, making it possible to sharpen images taken under adverse conditions.
Technology to Ensure Visibility of Objects at Night, in Bad Weather, and at Distance
Also, when taking a picture of a distant object with digital zoom function, which cuts out a portion of the picture and enlarges it, the number of pixels decreases, resulting in lower resolution. For this reason, Canon has also developed a technology called "super-resolution" that uses deep learning to complement image data to sharpen images. Applying deep learning technology, subjects, such as persons or automobiles, at a long distance can be more clearly recognized by estimating the true shape of the subjects and sharpening the captured images of them with degraded apparent resolution due to the optical characteristics of lens.
In addition, Canon has developed technologies to eliminate the effects of fog and haze for monitoring in difficult shooting environments. Fog and haze reduce the difference between light and dark. Thus, the resulting image loses contrast. The contrast depends on the distance to the subject. If a uniform contrast correction is applied to a shot image, the correction will be excessive for images that were relatively clearly visible, but insufficient for images that were not clearly visible. Therefore, Canon has developed a technology to correct image data appropriately by calculating the distance to the object to be photographed and the amount of fog and haze in the distance.Thanks to this technology, now we can reliably visualize what we want to see even at night, in bad weather, or in other situations where it is difficult for the human eye to see.
Images with reduced sharpness due to fog or haze (left) and images with sharpening processing (right)
Canon will continue to contribute to solving various social issues, by creating clear images beyond human vision, using image processing technology, which takes advantage of our own development and production of key devices such as lenses and sensors.
