Techniques for rate-distortion optimization in video compression

A device to compress motion video comprising:

• a processor circuit;
• a memory storing instructions which when executed by the processor circuit cause the processor circuit to:
• divide an unquantized coefficient of an unquantized coefficient block of a first frame of a video by a quantization value to derive a quotient, and to add a rounding value to and truncate the quotient to derive a quantized coefficient of a quantized coefficient block that corresponds to the unquantized coefficient block;
• receive, from a context-adaptive binary arithmetic coder (CABAC), an indication of a bitcost of a video bitstream resulting from compressing the video, the primary quantizer to alter the rounding in response to the indication;
• derive, based on a context model of the CABAC, a first bitcost value of using the quantized coefficient in compressing the video, the context model retrieved from among multiple context models based on an indication provided to the CABAC of the context model as an initial context model;
• derive, based on the context model of the CABAC, a second bitcost value of replacing the quantized coefficient with an alternate value derived by subtracting a value of one from an absolute value of the quantized coefficient in compressing the video;
• derive a first bitcost value and a second bitcost value for a second frame of the video as estimated bitcost values using the context model of the CABAC without incurring a delay to receive a new context model of the multiple context models as a new initial context model and without the CABAC generating actual bitcost values for the second frame, the first bitcost value for the second frame derived using a quantized coefficient of a quantized coefficient block of the second frame, the second bitcost value to replace the quantized coefficient of the quantized coefficient block of the second frame with an alternate value derived from quantized coefficient of the quantized coefficient block of the second frame;
• and
• store source data representing the video in a storage.

A computer-implemented method for compressing motion video comprising:

• deriving a first bitcost value of using a quantized coefficient of a quantized coefficient block of a first frame of a video in compressing the video, the first bitcost value based on a context model of a context-adaptive binary arithmetic coder (CABAC), the context model retrieved from among multiple context models based on an indication provided to the CABAC of the context model as an initial context model;
• concurrent with the derivation of the first bitcost value, deriving a second bitcost value of replacing the quantized coefficient with an alternate value derived by subtracting a value of one from an absolute value of the quantized coefficient in compressing the video, the second bitcost value based on the context model;
• selectively replacing the quantized coefficient with the alternate value in compressing the video based on a comparison of at least the first bitcost value and the second bitcost value;
• and
• deriving, based on the context model, a first bitcost value and a second bitcost value for a second frame of the video as estimated bitcost values without incurring a delay to receive a new context model of the multiple context models as a new initial context model and without the CABAC generating actual bitcost values for the second frame, the first bitcost value for the second frame derived using a quantized coefficient of a quantized coefficient block of the second frame, the second bitcost value to replace the quantized coefficient of the quantized coefficient block of the second frame with an alternate value derived from quantized coefficient of the quantized coefficient block of the second frame.

At least one non-transitory machine-readable storage medium comprising instructions that when executed by a computing device, cause the computing device to:

• derive a first bitcost value of using a quantized coefficient of a quantized coefficient block of a first frame of a video in compressing the video, the first bitcost value based on a context model of a context-adaptive binary arithmetic coder (CABAC), the context model retrieved from among multiple context models based on an indication provided to the CABAC of the context model as an initial context model;
• concurrent with the derivation of the first bitcost value, derive a second bitcost value of replacing the quantized coefficient with an alternate value derived by subtracting a value of one from an absolute value of the quantized coefficient in compressing the video, the second bitcost value based on the context model;
• selectively replace the quantized coefficient with the alternate value in compressing the video based on a comparison of at least the first bitcost value and the second bitcost value;
• and
• derive a first bitcost value and a second bitcost value for a second frame of the video as estimated bitcost values using the context model of the CABAC without incurring a delay to receive a new context model of the multiple context models as a new initial context model and without the CABAC generating actual bitcost values for the second frame, the first bitcost value for the second frame derived using a quantized coefficient of a quantized coefficient block of the second frame, the second bitcost value to replace the quantized coefficient of the quantized coefficient block of the second frame with an alternate value derived from quantized coefficient of the quantized coefficient block of the second frame.