Television Coding Using Two-Dimensional Spatial Prediction

Intensity correlation in television pictures extends both horizontally and vertically. 1-3 For a number of good reasons, most practical predictive encoders have exploited only the horizontal correlation: 3 (i) In a noninterlace system, access to the vertically adjacent picture element (pel) requires storage of one line of information (about 8,000 bits for broadcast TV or 2,000 bits for Picture1049 1050 THE BELL SYSTEM TECHNICAL JOURNAL, MARCH 1971 phone® Service). In a 2 : 1 interlaced system, a field plus a line must be stored (106 bits for broadcast or 250,000 for Picturephone Service). If access to the vertically adjacent pel in the same field is desired, a line of storage is required, (it) Most television systems, including Picturephone Service, use an interlaced scan. Because of the storage problem, access to the vertically adjacent pel in the same field is most practical. However, this pel is about twice as far away (spatially) from the current pel as is the previous pel in the same scan line. The vertical correlation of intensity within a field is thus less than the horizontal correlation. (in) Statistical studies of television pictures suggested that for linear predictive encoding very little extra redundancy could be removed by exploiting the vertical correlation as well as the horizontal correlation. 2,3 Given these reasons, why look at the use of vertically adjacent pels in various predictive encoding schemes? First, in recent years the cost of storage has fallen dramatically and hence the extra memory required for storing one line is not a serious objection.