Kansai Factors in Talking Image Effect of Digital Compressed Images

本文对数字压缩视频图像的说话图像效果进行了评价。采用MPEG-1格式对视频图像进行压缩，并引入劣等语音识别方法进行评价。用MPEG-1编码器对说话人的正面图像和侧面图像进行记录和压缩。视频比特率是在编码时设置的参数。采用较高的视频比特率进行编码，会导致存储空间中文件的体积增大，但却能得到高质量的视频图像。在本研究的实验中，分别使用400khs和4000kbs作为低视频比特率和高视频比特率。通过这个实验，下列事情得到了澄清。从TIE的角度来看，高视频比特率压缩与低视频比特率压缩没有显著差异。此外，对错误答案的分析也表明，视频比特率没有显著差异。侧边人脸图像的TIE高于正面人脸图像的TIE。由于说话人的视频图像是在彩蓝色背景下录制的，因此每个图像都具有很高的空间和时间冗余。由于该算法对MPEG-1编码的特点，可以很容易地将上述高冗余视频图像压缩成高度压缩的MPEG-1视频图像。因此，不同的视频比特率可能对图像效果没有影响。为了使用更复杂的视频图像，即时间和空间冗余较少的图像，引入了一种与CG图像重叠的说话视频图像。使用该视频图像发现，低视频码率下基于说话人嘴部运动的慢速效果比高视频码率下的慢速效果损失更大。

Talking Image Effect (TIE) of digital compressed video images was evaluated in this research. MPEG-1 format was used for video image compression and inferior speech recognition method was introduced for the evaluation.Talker's front face image and profile one were recorded and compressed with MPEG-1 encoder. Video bit rate is a parameter that is set at encoding. Encoding with higher video bit rate causes to make bigger file on the storage but it is expected to make higher quality video image. For experiment of this research, 400khs and 4000kbs were used as low video bit rate and high one respectively.Following things were clarified with this experiment.1. From the point of view of TIE, there was no significant difference between high video bit rate compression and low one. Furthermore, analysis on incorrect answer also indicated that there was no significant difference with video bit rate.2. TIE of profile face image is higher than that of front face image for labials.Because the talker's video image was recorded with a chromakey blue backdrop, each image had high spatial and time redundancy. Due to the feature of the algorithm on MPEG-1 encoding, it is easy to make a highly compressed MPEG-1 video image from above high redundant video images. Therefore there might be no difference of image effect with difference of video bit rate.In order to use more complex video image, that is a image with less redundant of time and space, a talking video image overlapped on the CG images was introduced. With using this video image, it was found that the slow effect based on motion of talker's mouth at low video bit rates was much loss than that of at high video bit rate.