Objective Quality Analysis of MPEG-1, MPEGZ & Windows Media Video

Transcription

1 Objective Quality Analysis of MPEG-1, MPEGZ & Windows Media Video Praveen K. Aeluri, Vinodh Bojan, Samuel Richie, Arthur Weeks richietilmail. ucfedu, weeksamail. ucfedu Abstract Video quality estimation mefhods in the literature can be broadly classified into 2 classes: objective and subjective. The metrics obtained from both classes of methods represent different intrinsic features of a video []. However, in this paper an objective qualit@ evaluatiort was performed 011 various videos, u,hich range fioni no motion 10 l7igh 177otion. encoded by three encoders at WO d@erentpunie rates. tu~ofianie sizes and three low bit rate combinations for fhe purpose of determining the change in video content introduced by the encoding and decoding process. The rflect of the encoding, bit rates, frame rates and frame sizes on the quality of videos as defined below has been studied. The mean-square-error based quality metvic, cumulative brightness error, used by this paper will enable comparison of video formats based on coiitent di/ferences and encoding parameter values. 1. ntroduction Conipression algorithms have made it possible to deliver various multimedia services over handwidthconstrained wired and wireless networks. t plays a vital role especially when running applications in real time, thus reducing the strain on streaming buffers. The availability of several complementary or competing compression formats gives rise to the determination of which parameters for encoding and decoding a video are best suited for a particular application. A study of the quality of several compressed video formats is pertinent []. n this paper. the quality of different video formats (MPEG, MPEG2 and Windows Media Formats) has been analyzed at different frame rates (1 Sfps and 3Ofps), frame sizes (320x240 and 640x480) and bit rates (48k, look and 300k), which are the most commonly found on nternet, on various motion types encoded in those formats. The sequences are 30 seconds in length and were selected from various movie clippings, television footage and generic video material to lepresent broad range of video content. 2. Digital Video Formats 2.1. Video Formats The MPEG formats are by far the most popular standard in recent times. n order to attain widespread use, the MPEG standard only specifies a data model for the compression of moving pictures and for audio signals. n this way, MPEG remains platform independent. There are primarily two processes that happen in the compression algorithm: spatial compression and temporal cornpression [2]. The extent of compression is governed by the psychovisual characteristics of human senses. n this way, the algorithm attempts to make use of the human inability to detect the absence of certain signals at any given point of time, to generate media tiles with significantly reduced file sizes. n this calculated method to exploit the limitations of human senses, a significant volume of original data and image information is lost. The extent of data loss depends on the degree of compression and the &ta rate of the system process. Quality comparison study is done by quantizing the quality using metrics. Objective metrics reveal the change in content and subjective nietrics characterize the degree of acceptable performance. Research on improving the quantitative and qualitative properties of digital video has focused predominantly on high bit rate video compression [2]. The underlying reason for the quality of a low hit rate video being degraded by the appearance of artifacts is still only vaguely understood [3]. An undentanding of how these defects are perceived, will lead to both higher quality low bit rate video compression techniques and metrics with which the quality of low bit rate video codes can be analyzed objectively. This paper does not provide a new metric to determine the quality, but discusses how one such metric, meansquare-rror (MSE), relates to the deviation from the original media material /04/% EEE 22 1

2 2.2 mage Quality Metric The objective metric used in this paper to determine the quality of the encoding and decoding methods is called he Cumulative Brightness Error (CBE) which is simply the cumulative mean square error of each frame s color components, averaged for the entire video sequence. t is hereafter just referred to as the Error, and is given by the following equation where R, G and B represent the values of Red, Green and Blue values of the pixels. 1 - the uncompressed AV1 and 2 - the compressed format.n. rn is the frame resolution, N is the frame number, and F is the number of frames in the video. 3. Experimental Analysis 3.1. Video Capture, Conversion and Frame Extraction The uncompressed AV1 video tiles are captured by a Viewcast Osprey@ Video Card at two different frame rates, 15 fps and 30 fps and two different frame sizes. 320 x 240 and 640 x 480. The raw AV1 files are then converted into MPEG-, MPEG-2 and Windows Media Video using TMPG EncodelB from Pegasys nc. and Windows Movie Maker@ respectively. Data rates of 48kbps, OOkbps and 300kbps, frame rates of Sfps and 30 fps and fame sizes of 320 x 240 and 640x 480 are chosen for conversion to refle& specifications of commonly found videos available on the nternet. The clips considered for analysis are of 30 seconds duration. Adobe Premiere@ is used to extract frames from the AV, MPEG-,.MPEG-2 and Windows Media Video files Test Data n order to analyze the performances of compression formats over a broad range of video content, four different kinds of video sequences are chosen depending upon the range..of motion of the objects in the video. The four video sequences chosen are as follows Animation. Anitnation clip is obtained from a an NTSC format broadcast source. Most of the frames have the same background with changes in the foreground Still Frames. A video with absolutely no motion was tested.the goal of this set up was to study how the conversion in the formats affects he quality of the video with no motion. So a picture in BMP format was converted to an uncompressed AV1 video Talking Head. A university lecture source video, which has a talking head with a steady background, was taken for analysis from a Florida Engineering Educational Delivery System (FEEDS) course. FEEDS is the distance education learning program in Engineering offered by the University of Central Florida Slides. This is also a UCF FEEDS video with a Microsoft Powerpoint slide presentation. 33. Quality Assessment The frames extractc:d from each video tile are stored in uncompressed BMP formats so that there b no loss in the detail in each frame in the process of extraction. Each pixel has its own RGB values depending upon the detail of that pixel. The RGB values from the pixels of frames from the converted tiles are then subtracted From the RGB values from the pixels of frames from thf: uncompressed AV1 tile. This gives us the error in each corresponding pixel in each corresponding frames from the two videos. The errors are then squared and added up for each frame and for entire duration of video clip. The errors are then divided by the total number of frames in the video clip and the total number of pixels depending on the frame size (320x240 or 640x480). The value thus obtained is called the Cumulative Brightness Error (CBE) which is simply called the error, hereafter as defined in the above formula Comparing MPEGl and MPEG2 The following tables show the Errors for all the video types encoded under MPEG- and MPEG-2 under different encoding conditions by varying frame sizes and bit rates. The errors have been calculated using the equation []. 222

3 Table 1: Analysis of MPEG- at 320 x 240 Table 4: Analysis of MPEG-2 at 640 x 480 M c P. Table 2: Analysis of MPEG -2, at 320 x fl" Fnmrr Slider 3.5. Data Frames Figure 1.a Table 3: Analysis of MPEG- at 640 x w - J8Lbls - A PowerPoin, Slidrr W , s Figure 1.b (b) Corresponding MPEG-2 Frame CBE= 640x fpsl 300 Kbps 223

4 Figure 2.a Figure 3.a i..,... ~., il.,...,... " * ~...,,..... Figure 2.b (b) Corresponding MPEG-2 Frame CBE= @ 640x480/ 30 fpsl300 Kbps 3.6. Analysis of WMV The WMV tiles were encoded using Windows Movie Maker. Due to limited options for choosing the frame rate, frame size and bit rate, very few combinations were obtained. 15 rpr 30 w J8 kw W bs JW L%p hi ma ti om Still Fnma SJ.615 9S Figure 3.b (b) Corresponding WMV Frame CBE= 320>:240/ 30 fps / 300 Kbps The data frames shown in this paper are the ones which exhibited a clear deviation from the original media file. 4. Conclusions Based on the error values obtained from the experimental set up, the following observations have been made: Animation video yielded the highest error in WMV. News, FEEDS - Talking Head, FEEDS -- PowerPoint slides and Still Frames performed significantly better in WMV than both MPEG codecs. Among the MPEG standards, MPEG- performed better overall than the MPEG-2. Given the results obtained in our experimental analysis, it can be assessed that 224

5 WMVis better suited for slow to no motion MPEG- works better for videos with high motion This difference in perfomnce can only be attributed to the intrinsic features and algorithm of each video encoding scheme. Each encoding scheme has its own measure ofpsycho visual tolerance. 5. References [] T.Eude, A.Mayache, An evaluation of Quality Metrics for compressed images based on Human Visual Sensitivity. Proc.lCSP 98, Beijing, China. [2] SO-EC/JTC/SC2/WGl, MPEG-Video Committee Draft [3] EePing Ong, Weisi Lin. Zhongkang Lu, Susu Yao, Xiaokang Yang and Fulvio Moschetti., Low Bit rate Video Quality Assessment based on perceptual characteristics, Proc. nternational Conference on mage Processing, O3.,V013, pp