Patent №
US 5,969,764
Granted
1999-10-19
Filed 1997
Owner
—
Lab
—
AI components
1
vision
Assignment
None on record
Dataset
AIPD
2023_r1 edition
Application
08896867
A method adaptively encodes a sequence of frames including video objects to provide a compressed video signal. The encoding is via a buffer having a ariable input rate and a constant output rate. The encoding uses a discrete cosine transform to produce coefficients that are quantized to generate image-representative code bits at a variable rate and texture, and motion and shape information for each video object stored in the buffer. The content of the buffer is restricted by adjusting quantization parameters with respect to a reference value and a quadratic rate distortion model to increase or decrease the number of bits stored in the buffer. Furthermore, the target number of bits for encoding each video object is estimated in accordance with a function of relative motion and size. The encoding bit rate is set to avoid buffer overflow.