Image Activity Characteristics in Broadcast Television

Abstract

Real-time processing of video waveforms has been implemented for the purpose of analyzing image dynamics. A digital preprocessor yields data sets that describe the difference between selected fields of an input video signal in both the spatial and temporal domains. Further reduction and analysis of the acquired data can provide several measures of image activity such as the frame-to-frame correlation, the framedifference signal (FDS), and statistics of scene lengths. The results of a series of measurements of the dynamics of typical broadcast television programs are presented. It is demonstrated that image activity is separable into two distinct classes, one caused by image and camera movement, and the other by scene switches.The oft-repeated television image is highly redundant and it is possible to exploit interframe and intraframe relationships of the image sequences to reduce the amount of information that must be transmitted.If it were not for the effect that flicker has on the human reception system, many video .communication processes could function adequately with image renewal rates well below 30 frames/s, since for a large class of television programs image motion is very slow. Such conditions often exist during teleconferencing and teleteaching. In these cases, activity, as measured by the change in the image, may be used as a trigger to initiate the renewal of the stored image portion of an audiovisual communication. For example, when a single television channel exists for the communication of a number of television images, access to the channel could be determined by the activity in each, and a new frame transmitted to a storage device associated with each source.The purpose of the work described in this correspondence was t o develop apparatus and procedures to measure the image activity of television programs in real time and to explore the hypothesis that these measures could be used to classify programs on the basis of the average number of frames required to adequately represent their content. The primary objective has been to obtain a small set of parameters with which to characterize activity.Although the redundancy in television images has been the subject of extensive studies related to potential bandwidth compression schemes, these studies have usually been restricted to simple image situations or to the analysis of limited amounts of data.In pioneering studies, Seyler [ 1 ] , [ 21 measured the probability distribution of frame-difference signals, and showed that Paper approved by the Editor for Communication Theory of the IEEE Communications Society for publication without oral presentation. Manuscript is with Computing Devices of Canada, Ottawa, Ont., Canada.the gamma distribution provides a good fit to the probability density function of television frame differences. This work has been repeated recently and similar results were reported Soroka [4] computed the interframe correlation of a sample television program by optically correlating a sequence of images that had been transferred on...

Separability of spatiotemporal spectra of image sequences

“…This finding is consistent with the applicability of the models of Eqs. 6 and 7 in earlier studies of spatial and temporal statistics [2,5,7,8,9].…”

“…An exponential autocorrelation function has been shown to be a good model for temporal frame to frame correlations of image sequences [e.g., 5,6,7,8], and for spatial correlations within each frame [e.g., 2, 3, 9].…”

Shot-boundary detection: unraveled and resolved?

“…In order to do so, suitable probability distribution needs to be found first. Studies reported in [22] and [7], involving statistical measurements of shot lengths for a large number of motion pictures, have shown that the distribution of shot lengths for all the films considered matches the Poisson function well [21]. Therefore, we adopt the Poisson function as the base for computing the cumulative probability .…”

“…Since the second item requires that the function converges toward the value 0.5, cumulative probability values need to be halved in order to ensure proper convergence, that is (21) The parameter of the Poisson distribution in (21) represents the average shot length of a video sequence, is the frame counter, which is reset each time a shot boundary is detected, and is the current shot length at the frame . The Poisson function was obtained in [7] and [22] as most suitable for motion pictures. However, we assume that the conclusions on shot-length distributions made by Salt and Coll can be extended further to all other types of video programs.…”

Fuzzy framework for unsupervised video content characterization and shot classification

“…26 Further statistical analysis can also yield valuable clues for film genre detection and style characterization. 19,27,28 For our application, we express the duration dur s of a shot s in minutes; i.e., dur s ϭ Number of frames in s 60•frame rate ͑fps͒ . ͑3͒…”