Method and apparatus for modifying a video stream to encode metadata

#This invention concerns the transmission of metadata related to an image sequence.

#It is very often necessary to associate data with a motion-image stream. In television so-called ‘metadata’ is frequently associated with one or more fields or frames of analogue or digital video. Typically the data associated with a field or frame describes parameters of that field or frame. Examples of parameters that can be carried by metadata include: identification of the video; timecode or frame number; the intended display aspect-ratio; or, a ‘signature’ of the field or frame that enables comparison with fields or frames in another video stream. For convenience, the term “video image” will sometimes be used in this specification to mean either video frame or video field.

#The association between the metadata and its respective image is frequently achieved by carrying the metadata in the blanking intervals of television signals, but, because these intervals are discarded by some video processing equipment, the metadata can be lost, or delayed differently from the video itself. This difficulty can be avoided by encoding the metadata within the active area of the video frames; this technique is known as ‘watermarking’. Typical watermarks are intended to be imperceptible to a viewer; known watermarking algorithms include methods that modify the frequency spectrum of the video, and methods that add low amplitude data signals to the video signal.

#FIG. 1 shows a block diagram of a metadata encoding process according to an example of the invention.

#FIG. 2 shows a block diagram of a metadata decoding process according to an example of the invention.

#FIG. 3 shows a block diagram of a system for encoding metadata in the pixels of an image.

#A common video process that is particularly damaging to watermarks is standards conversion, particularly frame-rate conversion. A typical frame-rate converter generates new fields or frames from input fields or frames by means of a temporal filter. Usually the filter forms each new field or frame from a weighted sum of input fields or frames. The weights are determined according to a filter aperture function that defines the relationship between the weight to be applied, and the time of the output field or frame with respect to the timing of the contributing input fields or frames. This time relationship is usually expressed as a temporal phase value. Typical frame-rate converters working between interlaced video formats combine four interlaced input fields to generate an output field. That is to say the filter aperture is four fields wide.

#The inventor has appreciated that, in a field-rate conversion process, the temporal phase of the conversion filter will vary cyclically, so that, at regular intervals, one input field or frame will provide the dominant contribution to the converter's output. For example, when the time of a particular input interlaced field coincides with the time of an output interlaced field there will be substantially zero contribution to the output from earlier or later input fields. This means that the impairment of a watermark due to standards conversion will vary cyclically, and substantially unimpaired data will be available at regular intervals.

#The time period of the cyclic variation corresponds to the frequency difference between input and output field rates. For example, when converting between nominal 60 Hz and nominal 50 Hz field rates, the temporal phase falls to a minimum value every six 60 Hz fields, or every five 50 Hz fields. And, when converting between 24 Hz frames and 25 Hz frames, the temporal phase falls to a minimum value every twenty-four 24 Hz fields, or every twenty-five 25 Hz fields. Typically the input and output rates of a video conversion process are not precisely ‘locked’ together and so the minimum temporal phase value that occurs at the difference frequency will, itself vary. There is no certainty that phase coincidence between input and output will occur. However, lower minimum values will occur when the frequency difference is small.