In recent years, to further improve image quality, non-Bayer image sensors have gradually become the go-to sensors (e.g., Quad, 2*2OCL, RGBW, etc.) in the mobile phone market. To adapt to various mainstream platforms (e.g., Qualcomm/MTK) and solve the image quality problems of such sensors, it is necessary to convert the RAW data of the sensors from a non-Bayer format to a standard Bayer format, which is a process called Remosaic.
Since the beginning of 2019, Tetras.AI has been developing an AI image-quality algorithm framework and IP solidification product system that is hardware-friendly on the sensor side. As an industry first, we've been able to solidify and embed the IP of the AI algorithm into an image sensor. The algorithm is designed to deeply integrate the various physical characteristics of the sensor and addresses the pain points present in a range of new image sensors. The algorithm offers high image quality, and is controllable, adjustable, and hardware-friendly. In addition, it can be customized to meet customer-specific application scenario requirements through different modes, such as Full-Resolution Mode, Binding Mode, and SF HDR (DOL) Mode.
Our AI-Remosaic technology comes in many product forms. It can be solidified to a sensor as an IP, run on the AP side as an SDK, and integrated into the AI-ISP chip as an IP. Currently, our AI-Remosaic technology can support a variety of sensors, including but not limited to Quad, 2*2OCL, 3*3, 4*4, 2*2*4OCL, and RGBW. There are already a number of new non-Bayer Sensor solidification products integrated with Tetras.AI's AI-Remosaic, which have been adopted by mobile phone customers.
Our AI Sensor provides a leading AI-based solution, adopts a unified framework, and allows for compatibility and adaptability with different CFA patterns. It not only offers image quality which is significantly better than traditional algorithm, but also affords intuitive and sufficient parameter configuration, which allows users to adjust the image quality according to fine-tuned parameters. Thanks to Tetras.AI's mature RTL design capability, we can ensure algorithm effectiveness while reducing IP size and power consumption to the greatest possible extent, and ensure massive performance with on-sensor solidification of the AI algorithm.