HiSilicon released latest image and video processing technology
HiSilicon has released latest image and video processing technology including Advanced DRC, Advanced WDR, 8-Stage 3DNR, CVBR (Constrained Variable Bit Rate) and Pre-filtering Encoding Technique.
Advanced DRC is the latest dynamic range compression technology. It comprehensively improves the dynamic range, enhances local contrast, and enhances the brightness of backlight face. It has great effect on wide dynamic scenes and infrared scenes.
At the same time, AE module provides the dynamic range value of scene, which can be used to judge the dynamic range of scene. Then sceneauto program can dynamically adjust DRC, AE, gamma and other parameters for better adaptability of scenes. The 3DNR refers to MG statistics for better denoising adaptability.
Advanced WDR is the latest WDR technology . It improves the detail extraction ability and wide dynamic brightness processing ability of DRC, thus realizing more compression of bright area and more enhancement of brightness of backlight area and more detail retention.
Advanced WDR combines ISP local contrast processing modules(Defog curve is used in WDR mode to suppress the brightness of high-brightness area, enhance local contrast.) and powerful 3DNR X,C interface. It makes the dynamic range significantly improved, the overall scene is clearer, more hierarchical. It can catch all the details you need in high dynamic range scenes.
8-Stage 3DNR is the latest denoising technology . It utilizes the 8-Stage filter denoising engine embedded in the chip. Compared with the general 4-Stage filter, it can realize the image denoising and enhancement simultaneously in the spatial-domain filtering process, as well as the blending of two different temporal modes in the temporal-domain filtering process. 8-Stage 3DNR also contains two IE modules for detail enhancement and one Sharpen module. Therefore, the 8-Stage 3NDR can achieve stronger denoising, stronger detail preservation and detail enhancement. Therefore, compared to our previous versions, the 8-Stage 3DNR is able to achieve clearer and sharper image quality in the static image area, on the other hand, it also has the capability to generate the clearer image quality for moving objects and in extreme low light conditions.
The temporal pumping artifact is a typical problem for video encoding which often occurred for IPC applications and seriously degrades the subjective video quality. The root cause of the pumping artifact lies in the video image quality discontinuity between I frames and their previous P frames.
Our new technique “Pre-filtering Encoding Technique” can significantly reduce or eliminate the artifact. By performing the block level-based temporal filtering for I frames, their subjective video quality is maintained consistent with their previous P frames, hence the entire video sequence is encoded with relatively stable quality so that pumping artifact is barely seen during the playback.
Constrained Variable Bit Rate (CVBR) is a rate control algorithm that minimizes the video quality fluctuation while conforming the target average bitrate and several instant bitrate constraints. In particular, these instant peak bitrate constraint and long-term average bitrate constraint guarantee the encoded bitstreams to be transmitted through a channel with a given bandwidth and/or stored in a device with specified volume or write speed. Under these constraints, an underlying Variable Bit Rate (VBR) control mechanism is employed to maximize picture quality over varying contents.
Figure 1 illustrate the PSNR, bitrate and file size performance for the CVBR compared against other rate control algorithms including VBR, AVBR and CBR. In this example, a video sequence comprising of segments with varying motion intensity is encoded using these algorithms. While encoding a low motion segment like at the beginning of the video, the CVBR naturally achieves low bitrate similar to that of VBR and AVBR, and hence saved some bits compared with CBR. When encoding a segment with intensive motion like at the end of the sequence, the CVBR would use some of the previously saved bit budget to allow instant bitrate go reasonably higher than the target average bitrate while still meeting the constraint of the overall bitrate or filesize. Therefore, compared with other rate control algorithms, the CVBR can reduce video quality fluctuation and hence achieve better video quality for the highly intensive motion segments. Meanwhile, it is worth noting that average bitrate from the CVBR is often lower than the target long-term bitrate which is the one from the CBR.