Çò̽ÌåÓý±È·Ö_Å·ÖÞ±­×ãÇòÍø£¤ÔÚÏßÖ±²¥

ͼƬ

TECHNOLOGY

Neural network technology in Canon apps

Find out all about Canon's AI-powered image processing technology and the benefits it brings when using image.canon and Digital Photo Professional.

After you've captured a photo in a split-second, it can then take hours to edit and perfect the resulting image. Wouldn't it be a great time-saver if an automated system could handle the initial steps for you, not just applying a standard formula as in-camera algorithms do but processing your RAW files individually to bring out the best in each one, based on an AI-powered analysis of its content? With Canon's Cloud RAW image processing service for image.canon, powered by its neural network tech, you can simply upload RAW files to the cloud directly from a compatible camera, and the neural network image processing system can intelligently sharpen details, reduce noise, and more.

The same technology also powers the Neural Network Image Processing Tool in Canon's Digital Photo Professional (DPP) software, intelligently analysing different areas within any given image, then applying a range of selective enhancements to achieve unprecedented levels of quality. Here we explain how it works.


Detail of an image shot in low light at high ISO of an aircraft flying against a dark blue sky, appearing very grainy and noisy.
 Detail from an image taken in low light at high ISO showing airport landing lights on elevated structures against a dark blue sky, with a lot of image grain and noise.

Cloud RAW image processing is particularly effective in retaining fine detail while suppressing image noise in night scenes, astrophotography and other situations where you're shooting under low lighting conditions with your camera set to a high ISO. The images on the left show the results you would typically expect in such conditions.

Detail of an image shot at high ISO of an aircraft flying against a dark blue sky, processed by image.canon's Cloud RAW image processing, so it appears sharper and less noisy.
Detail from an image taken in low light at high ISO showing airport landing lights on elevated structures against a dark blue sky, after processing by image.canon, with less image grain and noise.

"Improvements in fine detail and noise reduction can make shots taken at ISO 25,600 look like they were taken at ISO 6,400," says Canon Europe's Mike Burnhill. The result of neural network RAW image processing is evident in the cleaner, less grainy images on the right ¨C higher-quality images than in-camera RAW processing would normally produce.

What is neural network image processing?

Each shot you take is likely to be made up of disparate elements, such as a detailed landscape beneath a broad sky, or a portrait of somebody isolated against a defocused background. To realise the full potential of a RAW file, different areas within an image will benefit from a range of different treatments. That's where neural network image processing comes in, as Canon Europe Senior Product Marketing Specialist Mike Burnhill explains.

"Certain areas within an image can be highly detailed, like a portrait sitter's hair or a bird's feathers. Other areas might have very little detail, like the sky. Based on deep-learned AI, neural network RAW image processing can analyse an image, 'recognising' different types of subject matter in each specific area." The various areas are then processed individually.

"Areas of very fine detail and texture in an image can suffer from moir¨¦ patterning and false colour, but the system can eliminate these effects while also enhancing sharpness to really bring out the detail," Mike says. "Relatively undetailed areas don't require any sharpening, so none is applied, thus avoiding the risk of increasing image noise. It's like a painter using a variety of brushes for different areas of an image.

"Deep-learned AI processing sees the bigger picture. Instead of working on a pixel-by-pixel basis, it looks at larger areas within an image. It can pick out elements like diagonal lines and repeating shapes, then automatically apply the appropriate enhancements to refine levels of detail without making them look jagged or blocky. For noise reduction in either detailed or relatively plain areas, selective levels of luminance or chrominance noise reduction are automatically applied, to maximise fine detail and texture while minimising any graininess or colour speckling. In photographic terms, it equates to reducing the ISO setting by about two values without affecting the exposure."

A man taps on the image.canon option on his Canon camera's touchscreen to upload images and videos to the cloud, and a phone next to the camera shows the selected images.

image.canon simplifies the process of selecting, organising, sharing and transferring your images and videos. Simply register your camera, and you can then easily upload your images on the move. The service is free of charge, with only Cloud RAW image processing requiring a paid subscription.

A man looks through images uploaded to image.canon on a laptop with a Canon camera and a Canon SELPHY Square QX10 portable printer on the table alongside.

If you have a speedy internet connection, Mike says, the use of supercomputers for AI-powered Cloud RAW image processing can make it quicker to upload files to image.canon for processing than to apply the same neural network processing techniques on your own computer.

Cloud RAW image processing in image.canon ¨C linking your camera and transferring images

This neural network technology powers the Cloud RAW image processing service in image.canon. The service is based on a monthly subscription fee, covering 80 images a month. Bolt-ons are available for photographers who require greater capacity. The system will accept CR3 and CRAW files but doesn't handle RAW burst images. It's currently available for the Canon Çò̽ÌåÓý±È·Ö_Å·ÖÞ±­×ãÇòÍøÀÖ£¤ÔÚÏßÖ±²¥3, Çò̽ÌåÓý±È·Ö_Å·ÖÞ±­×ãÇòÍøÀÖ£¤ÔÚÏßÖ±²¥6 Mark II, Çò̽ÌåÓý±È·Ö_Å·ÖÞ±­×ãÇòÍøÀÖ£¤ÔÚÏßÖ±²¥7 and Çò̽ÌåÓý±È·Ö_Å·ÖÞ±­×ãÇòÍøÀÖ£¤ÔÚÏßÖ±²¥10 cameras, with more to come.

Linking your camera to image.canon is quick and easy. After you turn on your camera and select image.canon from the Wi-Fi/Bluetooth menu, a QR code will be displayed. Scan that QR code with your phone to be taken to image.canon. The first time you do this, the code will work to register your camera for the service.

Once you've registered and subscribed to Cloud RAW image processing, you can select images and upload them to the cloud for processing directly from your camera via Wi-Fi. To do this, go to the Cloud RAW image processing menu, select the capacity that suits your needs, click Add images to process and choose the images you want to send for processing. Once you've sent them, you'll be notified via email when your processed files are ready to download from image.canon. This means you can upload images while you're out on a shoot, and all the processing work can be done for you while you're on your way back to your home or studio.

A grainy and blurry image of an aircraft in flight against a red sky.

A low-light shot processed in-camera. Cloud RAW image processing does what in-camera RAW processing algorithms can't: it uses powerful AI with deep learning to analyse individual images' content and perform different processing in different areas to maximise image quality. The processed images are saved on image.canon. The original RAW files are stored alongside for 30 days and also remain on your camera as normal.

The same image of an aircraft in flight against a red sky, looking cleaner after processing using image.canon's Cloud RAW image processing.

A version of the same image, processed via image.canon to minimise noise and improve clarity. You can select your desired image output format before the image is uploaded to image.canon, including RAW, TIFF, JPEG and HEIF formats for stills.

Canon deep learning image processing technology

Canon has developed proprietary deep-learning image processing technology that enables the correction of image quality artefacts inherent to photography, and the upscaling of image files to larger resolutions with virtually no loss of quality.

Every moment is a slice of time ¨C a unique instant that will never occur again in the same way. Capturing these moments on camera is part of what makes photography so compelling ¨C whether it¡¯s breath-taking views seen for the first time, or intriguing events that tell a visual story.

To do justice to these points in time, we strive to maximise image quality and size ¨C overcoming natural phenomena such as digital noise, moir¨¦, softness from diffraction, and light fall-off towards the edge of the frame. We also want the most resolution possible in a photograph to maximise detail and give us the flexibility to crop into a scene without losing image quality.

Many of these phenomena are inherent to photography and can¡¯t be avoided even by the most skilled photographers. But now, Canon¡¯s advanced deep-learning software can correct the previously uncorrectable, tackling problems that are built-in to the principles of photography.

Our online Neural Network Image Processing Tool enhances and enlarges photographs using deep learning AI ¨C a technology inspired by the human brain that involves training a computer using large amounts of data .

Key optical factors that affect photos from 1 to 5

From the light source spectrum:

  1. Lens
  2. Infrared absorbent/anti-UV glass
  3. Low-pass filter
  4. Primary colour filter
  5. Shape of CMOS sensor opening

High resolutions, small files

The Neural network Upscaling Tool can enlarge images by up to four times while maintaining superb image quality, giving photographers the option of ultra-high resolutions without being weighed down by massive files in-camera.

That¡¯s ideal for sports and news photographers who value the competitive advantage of smaller image files when it comes to transmitting images remotely, but who might want to produce larger prints of final edited images, or crop into pictures for alternative compositions.

For example, the Neural network Upscaling Tool can upscale a 24-megapixel file to 96 megapixels, or a 45-megapixel file to a staggering 180 megapixels. Lower resolution images are transformed too, showing more definition and structure in the final result.

Our deep-learning engine works differently to conventional bicubic interpolation, interrogating the image and drawing on its knowledge of what different materials and textures look like, resulting in fewer artifacts and better fine detail. Unlike other AI upscaling tools, the Neural network Upscaling Tool uses camera and lens information and respects bokeh, preserving the background blur produced by shallow depth of field and producing a faithful result that is truthful to life.

GO LARGE! IMAGE UPSCALING, POWERED BY DEEP-LEARNING

This game-changing approach to high-resolution photography is available to all photographers ¨C not just those shooting with Canon. The tool runs locally on your Mac or PC (running 64-bit Windows 10 or Windows 11), featuring a simple user interface and compatibility with JPEG and TIFF files. It can also batch process files, saving you time in your digital workflow.

  • Try it now ¨C free for 30 days and just €5.94 per month after that. Discounts apply for a yearly subscription plan
  • Read more on the upscaling capabilities of Canon¡¯s Neural Network Image Processing Tool.

Organising your images

The image.canon service is also a smart choice for organising your image and video files and storing them safely in the cloud. Compatibility extends to all Canon Çò̽ÌåÓý±È·Ö_Å·ÖÞ±­×ãÇòÍøÀÖ£¤ÔÚÏßÖ±²¥ System cameras and the majority of DSLRs, including the EOS 6D Mark II, EOS 90D and EOS 850D, as well as a large number of EOS M series cameras and fixed-lens PowerShot and IXUS models.

"Connecting your Wi-Fi-compatible Canon camera enables you to seamlessly upload your images and movies in any file format, where they'll be stored in the cloud for 30 days," says Mike. "After that, lower-resolution thumbnails are retained, with 10GB of available storage space to accommodate even very large collections. Once they're in the cloud, you can access your original files from Canon's dedicated app or any web browser during the initial month, as well as being able to automatically forward them to your macOS or Windows computer and mobile devices.

"To extend connectivity and sharing options, image.canon can also connect to other cloud storage and social media services for long-term storage and sharing. These include Google Drive?, Google Photos?, Flickr, Instagram, Facebook and Twitter, as well as Adobe Photoshop Lightroom1."

Organisational tools enable you to create albums for easy access to collections of themed images. You can also sort images and videos based on individual cameras, file types and other parameters for intuitive cloud-based storing and sharing.

A before-and-after view of a photo of a church and gravestones being processed in the Digital Lens Optimizer in Canon's DPP software.

The Neural Network Image Processing Tool in DPP is able to analyse the image and apply more targeted corrections and enhancements. It can also save you time, making it ideal for busy professionals.

Neural Network Image Processing Tool in DPP

Digital Photo Professional (DPP), Canon's RAW processing and image editing software, is designed to make the most of RAW files from Canon cameras. The latest version introduces a powerful Neural Network Image Processing Tool, which is based on the same AI with deep-learning that powers the Cloud RAW image processing in image.canon.

Also a subscription-based service, the tool automatically analyses different areas within an image, and then applies selective enhancements based on neural network demosaic, lens optimisation and noise reduction algorithms.

The system also improves DPP's correction of lens imperfections. "We've digitally mapped our lenses for their behaviour at different apertures and focus distances, and for the different focal lengths of zoom lenses," Mike says. "But again, the AI in our neural networking system takes things to another level, analysing the image to see exactly what's happening and making the best possible enhancement. Along with enhanced sharpness and minimal noise, it can automatically correct chromatic aberrations, distortion and peripheral illumination."

The Neural Network Image Processing Tool has the added bonus of being able to automatically fix false-colour issues.

Frequently Asked Questions

Which platforms utilise Canon's neural network technology?

image.canon and DPP utilise Canon's neural network image processing to enhance photos.

Which cameras is Cloud RAW image processing available for?

The technology is available for Canon Çò̽ÌåÓý±È·Ö_Å·ÖÞ±­×ãÇòÍøÀÖ£¤ÔÚÏßÖ±²¥3, Çò̽ÌåÓý±È·Ö_Å·ÖÞ±­×ãÇòÍøÀÖ£¤ÔÚÏßÖ±²¥6 Mark II, Çò̽ÌåÓý±È·Ö_Å·ÖÞ±­×ãÇòÍøÀÖ£¤ÔÚÏßÖ±²¥7 and Çò̽ÌåÓý±È·Ö_Å·ÖÞ±­×ãÇòÍøÀÖ£¤ÔÚÏßÖ±²¥10 cameras, and is offered on a monthly paid-for subscription basis. This covers up to 80 images per month, with bolt-on extras for photographers who require processing of larger numbers of images.

What image formats can I send to image.canon?

All types of file formats for still images can be uploaded to image.canon, the choice of which is dictated by individual Canon cameras. These can include RAW, TIFF, JPEG and HEIF file formats for stills, as well as a variety of video file formats.

Can I download my images from image.canon?

Original, full-resolution files are retained in the cloud on image.canon for 30 days after they're initially uploaded. During this period, they can be downloaded to any computer or smart device. You can also set up automatic forwarding so that files are sent direct to your macOS or Windows computer, or to your mobile devices, or to external cloud services.

Which services is image.canon compatible with?

Image.canon is compatible with a wide range of cloud-based storage services and social media platforms. This enables you to transfer original image and movie files (depending on what each service supports) to other systems for permanent secure storage, as well as sharing photos and videos with family, friends and clients directly from image.canon, via your preferred social media platforms. The compatibility list includes the following:

Cloud-based storage services

  • Google Drive
  • Google Photos
  • Flickr
  • Adobe Photoshop Lightroom
  • YouTube

Social media platforms
  • Twitter
  • Facebook
  • Instagram

Find out more at the?image.canon?website.

1 Google, Google Photos and Google Drive are trademarks of Google LLC, and this site is not endorsed by or affiliated with Google in any way.

Adobe, Lightroom and Photoshop are either registered trademarks or trademarks of Adobe in the United States and/or other countries.

macOS is a trademark of Apple Inc., registered in the US and other countries and regions.

TWITTER, TWEET, RETWEET and the Twitter Bird logo are trademarks of Twitter Inc. or its affiliates.

Matthew Richards & Sarah Bakkland

Related articles

Get the newsletter

Click here to get inspiring stories and exciting news from Canon Europe Pro