Quantization device and quantization method
View Patent ↗A quantization device and quantization method are provided that reduce coding distortion with a small degree of calculation and achieve adequate coding performance thereby. A multistage vector quantization unit treats a number of candidates N that are designated prior to operation in the first-stage vector quantization unit, decrements the number of candidates by one beginning with the second-stage vector quantization unit and continuing with each stage thereafter. If the number of candidates is three or less, the multistage vector quantization unit assesses the quantization distortion at each stage, treating the number of candidates at the following stage as a predetermined value P if the quantization distortion is greater than a prescribed threshold, and treating the number of candidates at the following stage as a value Q that is less than the predetermined value P if the quantization distortion is less than or equal to the predetermined threshold.
1. A quantization apparatus that performs multiple-stage quantization by using tree search, comprising:
a searcher that matches each of one or more of encoding targets with a code vector stored in a codebook to select a certain number of candidates including one in order of smaller quantization distortion, the number of candidates being determined in a preceding stage or determined beforehand;
a calculator that calculates a quantization error vector by subtracting the code vector from the target for each of the candidates; and
a candidate number determiner that determines a certain number of candidates to be used in a subsequent stage, on the basis of the number of candidates determined in the preceding stage,
wherein, in the case where the number of candidates determined in the preceding stage is not greater than a value P specified beforehand, the candidate number determiner determines that, when a value of the quantization distortion is larger than a predetermined threshold value, the value P is used as the number of candidates in the subsequent stage, and determines that, when the value of the quantization distortion is not greater than the predetermined threshold value, a value Q smaller than the value P specified beforehand is used as the number of candidates in the subsequent stage.
2. A quantization apparatus that performs multiple-stage quantization by using tree search, comprising:
a searcher that matches each of one or more of encoding targets with a code vector stored in a codebook to select a certain number of candidates including one in order of smaller quantization distortion, the number of candidates being determined in a preceding stage or determined beforehand;
a calculator that calculates a quantization error vector by subtracting the code vector from the target for each of the candidates; and
a candidate number determiner that determines a certain number of candidates to be used in a subsequent stage, on the basis of the number of candidates determined in the preceding stage,
wherein, in the case where a present number of stages is at least a predetermined number of stages, or the number of candidates is not greater than a predetermined number of candidates P, the candidate number determiner determines that a predetermined number of candidates R is used in the subsequent stage, when a value of the quantization distortion is larger than a predetermined threshold value and also the number of candidates is smaller than the predetermined number of candidates R, and
the candidate number determiner determines that a predetermined number of candidates Q is used in the subsequent stage when the value of the quantization distortion is the predetermined threshold value or less and also the number of candidates is smaller than the predetermined number of candidates Q which is smaller than the number of candidates R.
3. A quantization method configured to perform multiple-stage quantization by using tree search, comprising:
matching each of one or more encoding targets with a code vector stored in a codebook to select, in a first stage, a certain number of candidates including one in order of smaller quantization distortion, the number of candidates being determined beforehand, and to select, in a second stage and in a stage subsequent to the second stage, a certain number of candidates in order of smaller quantization distortion, the number of candidates being determined in a preceding stage;
calculating a quantization error vector by subtracting the code vector from the target for each of the candidates; and
determining the number of candidates to be used in the subsequent stage, on the basis of the number of candidates determined in the preceding stage,
wherein, in the case where the number of candidates determined in the preceding stage is not greater than a value P specified beforehand, a candidate number determiner determines that, when a value of the quantization distortion is larger than a predetermined threshold value, the value P is used as the number of candidates in the subsequent stage, and determines that, when the value of the quantization distortion is not greater than the predetermined threshold value, a value Q smaller than the value P specified beforehand is used as the number of candidates in the subsequent stage.
4. A quantization method configured to perform multiple-stage quantization by using tree search, comprising:
matching each of one or more encoding targets with a code vector stored in a codebook to select, in a first stage, a certain number of candidates including one in order of smaller quantization distortion, the number of candidates being determined beforehand, and to select, in a second stage and in a stage subsequent to the second stage, a certain number of candidates in order of smaller quantization distortion, the number of candidates being determined in a preceding stage;
calculating a quantization error vector by subtracting the code vector from the target for each of the candidates; and
determining the number of candidates to be used in the subsequent stage, on the basis of the number of candidates determined in the preceding stage,
wherein, in the case where a present number of stages is at least a predetermined number of stages, or the number of candidates is not greater than a predetermined number of candidates P,
a candidate number determiner determines that a predetermined number of candidates R is used in the subsequent stage, when a value of the quantization distortion is larger than a predetermined threshold value and also the number of candidates is smaller than the predetermined number of candidates R, and
the candidate number determiner determines that a predetermined number of candidates Q is used in the subsequent stage when the value of the quantization distortion is the predetermined threshold value or less and also the number of candidates is smaller than the predetermined number of candidates Q which is smaller than the number of candidates R.