We describe a system for storing pictures in a database and retrieving them from a remote location over a low bit-rate channel. Even with compression techniques, it takes a long time to transmit a picture, making search for a picture of desired attributes time-consuming. Our system improves the human interaction with the picture database by constructing an auxiliary text database containing a list of attributes of each picture, a hierarchical encoder-decoder, and a light pen to select the areas of picture buildup. Picture searching (or browsing) takes place in two stages: in the first stage, knowing the required picture attributes, a user selects a subset of pictures by matching attributes to the text database; in the second stage, these selected pictures are displayed hierarchically so that a low resolution picture is reproduced first and made sharper gradually. A light pen allows the user to give priority to upgrade selected areas or reject a picture. Our techniques of hierarchical coding are simple to implement. Informal tests indicate that it is much easier to browse through a picture database using this system and that the time to retrieve a picture of given attributes decreases considerably compared to sequential picture presentation.

In processing video-telephone signals to reduce bandwidth requirements, lines may occasionally be lost due to buffer overflow or DPCM channel errors. The subjective degradation introduced by replacing deleted lines by averaging or repeating adjacent lines (substitution) has been measured. In the test, both skilled and unskilled observers were asked to add white noise to an unimpaired picture until the quality was equal to the same picture in which a number of lines had been randomly substituted. It was found that the more critical skilled observers deemed a picture, degraded by substituting 1 averaged line per frame, comparable to a signal-to-noise ratio of 38.6 dB, a rating of "definitely noticeable but not objectionable" on a 5-point impairment scale. Futhermore, it was found that repeating lines rather than averaging produced more degradation, equivalent to a 4 dB lower signal-to-noise ratio.