![Texas Instruments DM38x DaVinci Скачать руководство пользователя страница 185](http://html1.mh-extra.com/html/texas-instruments/dm38x-davinci/dm38x-davinci_user-manual_1097067185.webp)
Internal Modules
185
SPRUHI7A – December 2012 – Revised June 2016
Copyright © 2012–2016, Texas Instruments Incorporated
High-Definition Video Processing Subsystem (HDVPSS)
1.2.10 Noise Filter (NF)
1.2.10.1 Overview
Noise is one of the most important factors affecting video quality. Random noise can be sensor noise, A/D
conversion noise, analog transmission noise, etc. Since human eyes are very sensitive to oscillating
signals, the visual quality degenerates significantly even when the noise level is small. Video coding
efficiency also suffers from noise because the video encoder has trouble in matching noisy blocks and has
to spend extra bits to encode useless noise. Noise reduction plays a fundamental role in improving visual
quality and coding efficiency.
There are normally two types of noise reduction techniques – spatial filter and temporal filter. Spatial filter
processes each frame individually. Even though each frame becomes clean, the video may still look noisy
due to frame-to-frame intensity changing. Temporal filter can remove the frame-to-frame intensity
changing. But for moving areas, ghosting artifacts may occur because the pixel-to-pixel correspondence
between two consecutive frames is impacted due to the processing. In the noise reduction algorithm of the
NF module, a combination of spatial and temporal filter is applied to overcome the shortcomings of both
techniques.
It is desirable to have the spatial and temporal filters adaptive to the noise level of the video. If the filters
are too weak, the noise may be still obvious. If the filters are too strong, edges and details may be blurred
and ghosting artifacts may appear. Noise levels may vary a lot within a video sequence. Automated noise
estimation is necessary to measure the noise level of each video clip and adjust the parameters of the
filters properly.
This document describes the Video Noise Filter (NF) which is used to reduce noise in a video sequence in
order to improve video image quality and compression efficiency.
1.2.10.2 Features
The NF module has the following features:
•
Spatial and temporal video noise filtering
•
Built-in noise measurement
•
Adaptive to noise and motion
•
YCbCr 4:2:0 Video format support
1.2.10.3 Functional Description
The Video Noise Filter implemented in this device is a combination of spatial and temporal IIR filters. The
spatial filter is performed with neighboring pixels in the current frame while the temporal filter is performed
with a past filtered frame to suppress both spatial and temporal noises.
The spatial and temporal filters are adaptive to the noise level of the video to deal with varying noise
levels in different video contents. Noise estimation is performed by the hardware to measure the noise
level of each video clip and to automatically adjust the parameters of the filters properly.
shows the algorithmic block diagram of the noise filter.
The implementation of the video noise filter uses tile-based processing. The algorithm divides a frame into
32x32 tiles and process each of them independently. If the width or height of the frame is not a multiple of
32, 0x0s (for Luma) and 0x80s (for Chroma) will be padded into rightmost and/or bottommost tiles.