Method of Creating an Image Chain

Simple SummaryContent extracted from patent full text and abstract with AI.

This invention presents a method for building an 'image chain'—a sequence of steps to process images—where each traditional image processing function is replaced by an equivalent neural network. The sequence of neural networks can be trained together using backpropagation, significantly simplifying the parameter selection process and enabling more adaptive, optimizable, and user-friendly image processing pipelines. The approach can be used in imaging systems, such as those for medical radiology, to enhance image quality with less manual parameter tuning.

Use CasesContent extracted from patent full text and abstract with AI.

• Medical imaging systems (e.g., X-ray, CT, MRI) for image pre-processing and enhancement
• Industrial quality control imaging for automated defect detection and analysis
• Satellite and aerial image processing in geospatial analysis
• Photography and digital imaging applications for noise reduction and detail enhancement
• Security and surveillance applications requiring optimized image clarity and detection
• Microscopy and scientific imaging where image clarity and detail extraction are crucial
• Updating existing imaging software with AI-powered pipelines for improved performance without hardware changes

BenefitsContent extracted from patent full text and abstract with AI.

• Reduces the complexity of configuring and tuning parameters for each processing function, saving time and expertise required.
• Allows for joint optimization of all stages in the image processing chain, leading to potentially better overall image quality.
• Makes it easier for manufacturers and end-users to adapt and calibrate imaging devices to their specific requirements.
• Facilitates software-based updates and improvements to existing imaging systems, increasing flexibility and longevity of equipment.
• Streamlines the user experience by offering intuitive calibration steps where users can select their preferred output without needing deep technical knowledge.
• Potentially improves computational efficiency by collapsing or merging redundant neural network steps, reducing processing time and resource consumption.
• Supports both linear and non-linear image processing operations with neural networks, broadening the applicability of the approach.

Technical Classifications (CPCs)

Inventors & Applicants

Patent Abstract

The invention describes a method of creating an image chain (10), comprising the steps of identifying image processing functions (F1, F2, F3, F4) required by the image chain (10); replacing each image processing function (F1, F2, F3, F4) by a corresponding neural network (NN1, NN2, NN3, NN4); determining a sequence of execution of instances of the neural networks (NN1, NN2, NN3, NN4) to give the image chain (10); and applying backpropagation through the neural networks (NN1, NN2, NN3, NN4) of the image chain (10). The invention further describes an image processing method, and an imaging system (1).