Browsing by Author "Janse van Rensburg, Cornelius"

Now showing 1 - 1 of 1
  • Thumbnail Image
    Item
    Open Access
    Non-intrusive noise measuring on a television picture using spectrum analysis
    (1991) Janse van Rensburg, Cornelius; De Jager, Gerhard
    The objective of this thesis is to develop a new measuring technique to measure noise on a television picture non-intrusively. Existing methods measure noise intrusively by injecting test signals into the system or using techniques which make use of lines outside the picture area. From a system point of view one may assume no prior knowledge about the picture characteristics. The only knowledge assumed is that the noise is uniformly spread over the picture. Contrary to this, we propose that we do know the characteristics of a typical television picture through system and viewer constraints. Through spectrum analysis it will be proved that a television picture can indeed be characterized. For it to be acceptable to the general viewing public, it has to have a spectrum that is rapidly decreasing with increasing frequency. In fact, at the highest frequencies the picture spectrum should be virtually non-existent. On the other hand, the noise spectrum which has a flat spectrum over the whole frequency range, is most detectable at the high-frequencies. Thus, the high-frequency power of a picture spectrum can be used to estimate the noise power in the picture. Because the noise spectrum is not perfectly flat, a least squares error estimation is made by averaging the high-frequency spectrum points into one noise power estimate. This implies an assumed white noise spectrum. A method to make this noise estimation more robust against high-frequency harmonics, due to structure in the picture, is to tessellate the image into sub-images. This is done in order to obtain a statistical estimation of noise from the picture, because the picture spectrum estimation is a statistical estimation. All the noise estimations of all the sub-images are used to build a histogram to determine a noise estimate for the whole picture. The noise measuring technique, the Two Dimensional Noise Measuring Technique (2DSMT) was developed in Turbo Pascal (version 5) to run on an IBM PC. A minimum run-time of 50 seconds was obtained when run on a 25MHz 80386 machine with a numeric co-processor. A problem encountered was when composite video was digitized in the framegrabber. Due to the frequency division multiplexing of the PAL color system, the high-frequency power is not primarily due to noise, but due to color information. Therefore the 2DSMT only works on monochrome input, which may be a decoded red, green or blue signal. The 2DSMT compared favorably with the existing analogue noise measuring techniques where a constant luminance signal is used as input. Furthermore, it compared well with a noise measuring technique developed during the testing operation which measured constant luminance areas on cartoon pictures.