|
De-Interlacing The following clips illustrate the importance of proper de-interlacing when creating streaming files greater than one video field (240 lines) in height.Motion Adaptive de-interlacing is one of the most advanced forms of de-interlacing technology available. Motion adaptive de-interlacing combines the best aspects of both Bob (linear spatial) and Weave (linear temporal) technology by isolating the de-interlacing compensation to the pixel level. Where Bob/Weave and VT processing effect the entire frame of video, motion adaptive de-interlacing (as implemented on the DRC Stream hardware) processes each pixel independently, resulting in the highest quality image possible. Areas of no motion are statically meshed (weave) and areas where motion is detected are treated with a proprietary filtering technique resulting in very high quality, progressive scan images.Properly processing the video from its native interlace form factor to high quality progressive scan data is extremely important to the overall quality of the resulting image. Competitive hardware results.
Raw footage captured using the competitions hardware and encoded with all filtering options turned on. The footage is captured at full resolution as a demonstration of how interlaced material is handled by capture systems without advanced de-interlacing capabilities. The presence of the interlace artifacts causes many compression artifacts in the image, not the least of which is poor presentation quality. Pause the video just as the wheels bounce through the scene to see the full effect. Software based clip.
This is an example of the same footage captured as an uncompressed AVI file and processed using a very popular software solution. Notice that there are less artifacts overall and that the image quality is improved somewhat. The overall effect of the software based de-interlace engine is motion blur (trails) artifacts in areas of high motion. Pause the clip just as the wheels bounce across the screen to see the full effect. (software processing time was 40 minutes) StreamZ real-time results
Same footage captured using a Digital Rapids StreamZ media encoding server. This footage was captured and processed in real-time resulting in a high quality, progressive scan file. Notice the lack of interlace artifacts and the overall higher quality of the clip. Pause the video anywhere in the clip and notice that there are no interlace artifacts. Noise Reduction Typical 3D noise reduction as used by most products compares images in sequence over a predetermined period of time, blending the images between fields or frames. The simplified theory is that by blending data over time you can reduce the overall noise content of the resulting image. This is used in the film industry to remove film grain from blue or green screen shots. Successive samples of the same blue or green screen scene are blended together which will effectively cancel out the noise found in any individual frame. The problem with typical noise reduction algorithms is that if there is any motion in the individual scene you will get a motion blur (trails) artifact in the processed frame. One answer to this is to limit the blending process to a specific range of frames or to restart the process every few frames. This is very visible in the noise reduction found in the some software codecs. Every few frames it will reset and the noise will suddenly appear only to disappear several frames later, then the process starts over again. The DRC-Stream hardware uses a very advanced form of temporal noise processing called motion adaptive 3D noise reduction. Motion adaptive 3D noise reduction combines the information in multiple frames of video on a pixel by pixel basis to decide how much processing is applied to each pixel in the final frame. Because each pixel is calculated individually the result will be more precise noise reduction with less motion blurring artifacts than would otherwise be normally possible. Without noise reduction :
This clip represents the raw footage as captured by a 3 chip DV camera in very challenging lighting conditions. Obviously there are quite a number of things wrong with the footage in question. The video is too bright, there is a lot of video noise from the camera's auto gain (trying to compensate for low light) and there is very little contrast in the image. Look in the background of the clip and you will see that the graininess induced by the camera. With 3D motion adaptive noise reduction :
3D motion adaptive noise reduction has been applied to the clip and there is already a significant improvement in quality over the original footage. With 3D motion adaptive noise reduction and some simple processing applied :
Some simple proc amp adjustments have been applied to the clip to try and restore some contrast and saturation to the image. The footage is not perfect, but it is no longer wasted footage and can now probably be used for its original purpose. Comparison of various bitrates The following four clips were all encoded simultaneously, in real-time on a Digital Rapids StreamStation. These clips are not an example of the ultimate quality possible using a Digital Rapids solution but rather what is possible using consumer equipment in real world conditions. The camera used is a 10 year old Hi-8 camera, it is hand held and the lighting conditions vary from poor to terrible. Notice that the camera never stops moving through the clips creating a challenging situation for the codec. All of the clips are sized somewhat above what would be considered normal resolution for these bitrates. During the opening frames of the clip look at the shallow angle of the wires that make up the tiger cage. On a system without proper de-interlacing the wires would show up as jagged lines without the clear definition that you see in these clips. Zoo footage - 64Kb/s ,
128Kb/s ,
512Kb/s,
1Mb/s
MPEG-2 samples
Slime - same clips as used to demonstrate motion adaptive interlace above, encoded at ~2Mb/s MPEG-2 format and 640x480 resolution.
DHL - a commercial spot for DHL encoded at ~2Mb/s VBR MPEG-2 format at 640x480 resolution ZooTest - some amatuer footage shot with a consumer camera to show off 3D noise reduction and motion adaptive de-interlacing in a low bitrate MPEG-2 file at 640x480 resolution. ZooTest2 - more amatuer footage shot with a consumer camera to show off 3D noise reduction and motion adaptive de-interlacing in a low bitrate MPEG-2 file at 640x480 resolution.
MPEG-1 SamplesZooTest - some amatuer footage shot with a consumer camera to show off 3D noise reduction and motion adaptive de-interlacing in a low bitrate MPEG-1 file at 320x240 resolution.
ZooTest - High resolution - some amatuer footage shot with a consumer camera to show off 3D noise reduction and motion adaptive de-interlacing in a low bitrate MPEG-1 file at 640x480 resolution. DRC stream (PCI cards) 
|
