‘Raptor' includes tunable blocks to correct for various lens aberrations, for variations in illumination across the sensor surface, to remove noise introduced by very high density sensors in low light and to remove bad pixels. Then the raw image turned into an RGB image and tuned for the characteristics of the sensor before being sent off for display and/or encode. All of these operations are said to be done in the most configurable way possible in order to interface to the widest selection of CMOS sensors.
The job of the ISP is evolving rapidly away from being a simple image or video capture device; forward looking features of the ‘Raptor’ architecture include the ability to interact with other powerful vision and video processors in the multimedia subsystem. One example of this is a dedicated interface to a PowerVR video encoder in order to implement low latency video encode for live video conferencing; another is the production of image statistics, or hints, which can be used for a range of imaging and vision enhancements.
Computational photography using the GPU for heavy duty processing is already within the range of the abilities of ‘Raptor’, and forward looking applications include vision applications using multi-camera arrays (‘Raptor’ can simultaneously support multiple sensors with widely different characteristics) as well as higher pixel depth applications for industrial and automotive use. In its most cost-efficient form, ‘Raptor’ supports 10-bit pixel depths which matches the 10-bit encode and decode capability already offered by PowerVR video processors. It can be configured to support much higher bit depths (up to 16) when needed.
|PowerVR Vision Platform|