JP2004056605A - Method, device and program for controlling code amount - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide a code amount controlling method, a code amount controller and a code amount control program for calculating a bit position at which distortion quantity becomes minimum with a small amount of operation when an image is divided into blocks of a particular size and bit positions for discarding the bit plane of each of the blocks are determined in encoding the image under the upper limit value of a code amount. <P>SOLUTION: This code amount controller 3 is provided with a code block dividing means 10, a distortion/code amount table generating means 11, a code amount reference value list generating means 12, a work table generating means 13, a reverse searching means 14 and a bit omitting means 15. In addition, a bit position at which the total distortion quantity between a first block and a particular block is minimum and the total distortion amount at that time is generated as a work table for every value of a plurality of code amount reference values where a code amount upper limit value is an upper limit. and a bit position where the total distortion amount of the whole blocks is minimum is reversely searched from the work table. <P>COPYRIGHT: (C)2004,JPO

Description

【００９７】
この（８）式において、ａｒｇｕｍｅｎｔ｛条件｜パラメータリスト｝は「パラメータリスト」から「条件」を満たすパラメータを取り出す操作を表すものとし、「条件」における「≒」記号は「最も近い」を表す記号とする。
そして、このステップＳ４２のコードブロックＢ（１），…，Ｂ（ｊ）の歪総量ｆ（ｉ，ｊ；ｔ）の算出には、（９）式を用いる。
【００９８】

【００９９】
この（９）式において、（ｋ（ｉ）−Ｒ（ｔ，ｊ））は、（サブ問題Ａ）Ａ（ｉ，ｊ）に割り当てられた符号量基準値ｋ（ｉ）から、コードブロックＢ（ｊ）の符号量Ｒ（ｔ，ｊ）を引いた量で、ここでは残余符号量と呼ぶこととする。また、ψ（ｋ（ｉ）−Ｒ（ｔ，ｊ））は、（８）式に示したように、残余符号量（ｋ（ｉ）−Ｒ（ｔ，ｊ））に「最も近い」符号量基準値リストＫｓｅｔの要素のインデックスを表している。従って、歪総量テーブルＦのＦ（ψ（ｋ（ｉ）−Ｒ（ｔ，ｊ）），ｊ−１）は、（サブ問題Ａ）Ａ（ψ（ｋ（ｉ）−Ｒ（ｔ，ｊ），ｊ−１）を解いたときの歪総量となる。
【０１００】
そして、（９）式は、この（サブ問題Ａ）Ａ（ψ（ｋ（ｉ）−Ｒ（ｔ，ｊ），ｊ−１）を解いたときの歪総量に、コードブロックＢ（ｊ）の歪量Ｄ（ｔ，ｊ）を加算したもので、（サブ問題Ａ）Ａ（ｉ，ｊ）において、ＴＰのリストであるＴ（ｊ）の最後の要素ｔ（ｊ）をｔとしたときのコードブロックＢ（１），…，Ｂ（ｊ）の歪総量となる。
このように、（９）式で算出した歪総量が最小となるＴＰの値ｔを（１０）式を用いて取得する（ステップＳ４３）。
【０１０１】

【０１０２】
この（１０）式は、歪総量ｆ（ｉ，ｊ；ｔ）がパラメータｔによって変化するときに最小値（ｍｉｎｉｍｉｚｅ）となるｔの値を取り出すことを表している。ここで、歪総量ｆ（ｉ，ｊ；ｔ）の要素は最大Ｐ（０，…，Ｐ−１）個あり、その中の最小値が、（サブ問題Ａ）Ａ（ｉ，ｊ）を解いたときの歪総量である。従って、その歪総量が最小値となるときのＴＰの値ｔを取得すれば、このｔの値が、（サブ問題Ａ）Ａ（ｉ，ｊ）の解であるＴＰのリストＴ（ｊ）における最後の要素ｔ（ｊ）となる。
【０１０３】
そして、ステップＳ４３で取得したＴＰの値ｔを、インデックステーブルＩｄｘ（ｉ，ｊ）に格納し（ステップＳ４４）、そのときのｔの値における歪総量ｆ（ｉ，ｊ；ｔ）を、歪総量テーブルＦのＦ（ｉ，ｊ）に格納する（ステップＳ４５）。そして、変数ｉが符号量基準値の数（Ｓ（Ｋ））かどうかを判定し（ステップＳ４６）、異なる場合（Ｎｏ）は変数ｉの値を１加算して（ステップＳ４７）、ステップＳ４２へ戻って処理を続ける。
【０１０４】
そして、変数ｉが符号量基準値の数（Ｓ（Ｋ））と等しい場合（Ｙｅｓ）は、変数ｊがコードブロックの数（Ｎ）かどうかを判定し（ステップＳ４８）、等しい場合（Ｙｅｓ）は処理を終了し、異なる場合（Ｎｏ）は変数ｊの値を１加算して（ステップＳ４９）、ステップＳ４１へ戻って処理を続ける。
【０１０５】
以上の動作によって、インデックステーブルＩｄｘ及び歪総量テーブルＦにおける第２列目以降のコードブロックＢ（ｊ）（ｊ＝２，…，Ｎ）の値がＳ（Ｋ）個づつ順に埋められ、（サブ問題Ａ）Ａ（ｉ，ｊ）を解いたときの、インデックステーブルＩｄｘの値であるＩｄｘ（ｉ，ｊ）と、歪総量テーブルＦの値であるＦ（ｉ，ｊ）が決定される。このインデックステーブルＩｄｘ及び歪総量テーブルＦの要素が逐次的に決定される順序を図９に示す。図９（ａ）には、インデックステーブルＩｄｘの決定順序を、図９（ｂ）には、歪総量テーブルＦの決定順序を模式的に示している。インデックステーブルＩｄｘ及び歪総量テーブルＦ共に、第１列目（コードブロックＢ（１））の符号量基準値ｋ（ｉ）のインデックスｉ＝１からｉ＝Ｓ（Ｋ）までが初期値として設定された後に、第２列目のコードブロックＢ（２）から順番にコードブロックＢ（Ｎ）までの値が決定される。
【０１０６】
このように、帰納的に（サブ問題Ａ）を解いていくことで、（サブ問題Ａ）Ａ（Ｓ（Ｋ），Ｎ）における歪総量テーブルＦの値であるＦ（Ｓ（Ｋ），Ｎ）と、インデックステーブルＩｄｘの値であるＩｄｘ（Ｓ（Ｋ），Ｎ）とを求めることができる。
【０１０７】
［ステップＳ５：逆探索］
次に符号量制御装置３は、作業テーブル生成手段１３によって、ステップＳ４で設定されたインデックステーブルＩｄｘのＳ（Ｋ）行Ｎ列のテーブルであるＩｄｘ（ｉ，ｊ）（ｉ＝１，…，Ｓ（Ｋ），ｊ＝１，…，Ｎ）を用いて、Ｉｄｘ（Ｓ（Ｋ），Ｎ）、すなわち、コードブロックの数がＮ個の場合のコードブロックＢ（Ｎ）におけるＴＰの値ｔ（Ｎ）を初期値として、逆探索によりｔ（Ｎ−１），…，ｔ（１）を順次決定する。
【０１０８】
ここで、逆探索（ステップＳ５）の動作を図８のフローチャートを参照して説明する。
図８に示すように、逆探索（ステップＳ５）は、まず、符号量基準値ｋ（ｉ）のインデックスである変数ｉを符号量基準値の数（Ｓ（Ｋ））に初期化し（ステップＳ５０）、コードブロックＢ（ｊ）のインデックスである変数ｊをコードブロックの数（Ｎ）に初期化する（ステップＳ５１）。
【０１０９】
そして、コードブロックＢ（ｊ）に割り当てるＴＰの値であるＩｄｘ（ｉ，ｊ）をインデックステーブルＩｄｘから取得し、ｔ（ｊ）に設定する（ステップＳ５２）。このステップＳ５２は、あるＩｄｘ（ｉ，ｊ）が定まっているとすると、符号量基準値ｋ（ｉ）をコードブロックＢ（１），…，Ｂ（ｊ）に割り当てる（サブ問題Ａ）Ａ（ｉ，ｊ）はすでに解かれており、Ｉｄｘ（ｉ，ｊ）は、最後のコードブロックＢ（ｊ）に割り当てるＴＰの値であるｔ（ｊ）となる。
【０１１０】
この結果、符号量基準値ｋ（ｉ）から、コードブロックＢ（ｊ）に割り当てる符号量Ｒ（ｔ，ｊ）を引いた量、すなわち、残余符号量ｋ（ｉ）−Ｒ（ｔ（ｊ），ｊ）がコードブロックＢ（１），…，Ｂ（ｊ−１）に割り当てるべき符号量となる。そこで、このコードブロックＢ（ｊ）の残余符号量ｋ（ｉ）−Ｒ（ｔ（ｊ），ｊ）に「最も近い」符号量基準値のインデックスをψ（ｋ（ｉ）−Ｒ（ｔ（ｊ），ｊ））で算出し、新たに変数ｉに代入する（ステップＳ５３）。
【０１１１】
そして、コードブロックＢ（ｊ）のインデックスである変数ｊから１減算し（ステップＳ５４）、変数ｊが０になったかどうかを判定する（ステップＳ５５）。ここで変数ｊが０になった場合（Ｙｅｓ）は、処理を終了し、変数ｊが０でない場合（Ｎｏ）は、ステップＳ５２へ戻って処理を続ける。
【０１１２】
以上の動作を図１０を参照してさらに説明する。図１０は、逆探索手段１４における逆探索動作を模式的に表した図である。インデックステーブルＩｄｘの値であるＩｄｘ（Ｓ（Ｋ），Ｎ）、すなわちコードブロックＢ（Ｎ）のＴＰの値（ｔ（Ｎ））は、すでにステップＳ４によって求められている（Ａ１に書き込まれている）。次に、符号量上限値Ｋから、ｔ（Ｎ）でコードブロックＢ（Ｎ）を符号化したときの符号量（Ｒ（ｔ（Ｎ）、Ｎ））を引いた残余符号量（Ｋ−Ｒ（ｔ（Ｎ）、Ｎ））に「最も近い」符号量基準値リストＫｓｅｔの要素のインデックスをｉとする。すなわち、Ｋ−Ｒ（ｔ（Ｎ）、Ｎ）≒ｋ（ｉ）となるインデックスｉを求める。そして、インデックステーブルＩｄｘで、このインデックスｉに該当する（Ａ２に書き込まれている）コードブロックＢ（Ｎ−１）の値（Ｉｄｘ（ｉ，Ｎ−１））をコードブロックＢ（Ｎ−１）のＴＰの値（ｔ（Ｎ−１））とする。
【０１１３】
さらに、符号量基準値ｋ（ｉ）から、ｔ（Ｎ−１）でコードブロックＢ（Ｎ−１）を符号化したときの符号量（Ｒ（ｔ（Ｎ−１）、Ｎ−１））を引いた残余符号量（Ｋ−Ｒ（ｔ（Ｎ−１）、Ｎ−１））に「最も近い」符号量基準値リストＫｓｅｔの要素のインデックスをｉ´とする。すなわち、Ｋ−Ｒ（ｔ（Ｎ−１）、Ｎ−１）≒ｋ（ｉ´）となるインデックスｉ´を求める。そして、インデックステーブルＩｄｘで、このインデックスｉ´に該当する（Ａ３に書き込まれている）コードブロックＢ（Ｎ−２）の値（Ｉｄｘ（ｉ´，Ｎ−２））をコードブロックＢ（Ｎ−２）のＴＰの値（ｔ（Ｎ−２））とする。このように、順次ＴＰの値をｔ（１）まで逆探索する。
【０１１４】
以上の動作によって、Ｎ個のコードブロックＢ（ｊ）のＴＰのリストＴ＝（ｔ（１），…，ｔ（Ｎ））は、すでに求まっているインデックステーブルＩｄｘを参照（逆探索）することで、ｔ（Ｎ），…，ｔ（１）の順番に決定される。
以上のステップＳ１〜Ｓ５の動作により、与えられた符号量上限値Ｋの下で歪量総和Ｄ（Ｔ）（（５）式参照）を最小化するコードブロックＢ（ｊ）（ｊ＝１，…，Ｎ）についてのＴＰ（Ｔｒｕｎｃａｔｉｏｎ　Ｐｏｉｎｔ）のリスト（ＴＰリスト）Ｔを逐次的に決定することができる。
【０１１５】
このステップＳ１〜Ｓ５の動作で明らかなように、本手法によれば（問題Ａ）は、Ｓ（Ｋ）＊Ｎ＊Ｐのオーダの演算で解くことが可能である。例えば、符号量基準値の数を１００（Ｓ（Ｋ）＝１００）、コードブロックの数を６４（Ｎ＝６４）、各コードブロックのビットプレーンの数を１６（Ｐ＝１６）とすると、５１２００回の演算ステップで動作させることが可能で、例えば、１ステップあたり１００ナノ秒（ｎｓ）の演算時間としても、１／３０秒のフレームレートで本手順を動作させることができる。これにより、ソフトウェアによる符号量制御を実現することができる。
【０１１６】
なお、本手法において、量子化レベルをビットプレーンの位置を示すＴＰ（Ｔｒｕｎｃａｔｉｏｎ　Ｐｏｉｎｔ）として説明したが、この量子化レベルは、サブバンド符号化で使用されるビットプレーン符号化に限定されるものではなく、ブロック単位で符号量と、歪量とをパラメータ制御するような符号化方式においても、そのパラメータを量子化レベルとみなすことで実現することが可能である。例えば、ＤＣＴ（Ｄｉｓｃｒｅｔｅ　Ｃｏｓｉｎｅ　Ｔｒａｎｓｆｏｒｍ：離散コサイン変換）ベースのブロック符号化においては、ブロック単位に量子化レベルを決定すればよい。
【０１１７】
また、ＪＰＥＧ２０００では、１つのビットプレーンが３種類のパスに分類されてエントロピ符号化されるが、本手法においては、パス毎に別のビットプレーンとみなすことで前記した動作と同様に扱うことが可能である。
【０１１８】
【発明の効果】
以上説明したとおり、本発明に係る符号量制御方法、符号量制御装置及び符号量制御プログラムでは、以下に示す優れた効果を奏する。
【０１１９】
請求項１、請求項２、請求項４又は請求項５に記載の発明によれば、情報量を廃棄（量子化）することで、所定の符号量上限値の範囲内で圧縮符号化を行う画像符号化において、歪量を最小化する符号をテーブル検索及び四則演算による短いステップの演算で決定することができる。
これによって、汎用のＣＰＵであっても、動画像を実時間で高画質に圧縮符号化を行うことが可能になり、低コストで符号化装置を実現することができる。
【０１２０】
請求項３に記載の発明によれば、量子化を行うレベル（量子化レベル）をビットプレーン符号化におけるビットプレーンを廃棄する基準位置とすることで、例えば、ビットプレーン符号化によって量子化を行うＪＰＥＧ２０００等で、画像をブロックに分割し、そのブロックのビットプレーンを廃棄する位置を各ブロック毎に制御することが可能になる。
【図面の簡単な説明】
【図１】本発明の実施の形態に係る符号量制御装置の全体構成を示すブロック図である。
【図２】本発明の実施の形態に係る歪量テーブル及び符号量テーブルの構造を説明するための説明図である。
【図３】本発明の実施の形態に係る符号量基準値リストの構造を説明するための説明図である。
【図４】本発明の実施の形態に係る（問題Ａ）の解を求めるためにコードブロックを複数の対象に分解する作業を説明するための説明図である。
【図５】本発明の実施の形態に係る符号量制御動作の概略を示すフローチャートである。
【図６】本発明の実施の形態に係る作業テーブル初期設定の動作を示すフローチャートである。
【図７】本発明の実施の形態に係る作業テーブル設定の動作を示すフローチャートである。
【図８】本発明の実施の形態に係る逆探索の動作を示すフローチャートである。
【図９】本発明の実施の形態に係るインデックステーブル及び歪総量テーブルを設定する順序を視覚的に表した模式図である。
【図１０】本発明の実施の形態に係る逆探索動作を模式的に表した模式図である。
【図１１】サブバンド符号化におけるサブバンド分解とコードブロック分割を説明するための説明図である。
【図１２】ビットプレーン符号化の動作の概要を説明するための説明図である。
【図１３】ビットプレーン符号化におけるＴｒｕｎｃａｔｉｏｎ　ｐｏｉｎｔと符号量及び歪量との関係を説明するための説明図である。
【図１４】ビットプレーンの位置（大きさ）と、歪量及び符号量の関係を説明するための説明図である。
【符号の説明】
１……符号化装置
２……サブバンド分解手段
３……符号量制御装置
４……エントロピ符号生成手段
１０……コードブロック分割手段
１１……歪・符号量テーブル生成手段
１２……符号量基準値リスト生成手段
１３……作業テーブル生成手段
１４……逆探索手段
１５……ビット切り捨て手段
Ｄ…………歪量テーブル（歪・符号量テーブル保持手段）
Ｒ…………符号量テーブル（歪・符号量テーブル保持手段）
Ｋｓｅｔ…符号量基準値リスト（符号量基準値リスト保持手段）
Ｉｄｘ……インデックステーブル（作業テーブル保持手段）
Ｆ…………歪総量テーブル（作業テーブル保持手段）[0001]
TECHNICAL FIELD OF THE INVENTION
The present invention relates to an image coding technique, and more particularly, to compressing an image by truncating coded data under a condition where the amount of code is limited, and compressing the image with the least amount of distortion. TECHNICAL FIELD The present invention relates to a code amount control method for performing code amount control for performing coding, a code amount control device, and a code amount control program.
[0002]
[Prior art]
2. Description of the Related Art Conventionally, as an encoding method for compressing and encoding an image by truncating encoded data under a condition in which a code amount is limited, JPEG2000 (Joint Photographic Expert Group 2000) which performs encoding by wavelet transform. And the like.
[0003]
In JPEG2000, an input image is decomposed by two-dimensional subband decomposition using a wavelet transform, pixels of the decomposed subbands are represented by a plurality of bits (P bits), and each subband is spatially separated. Is divided into blocks (code blocks) having a specific size (vertical m pixels, horizontal n pixels). Then, P bit planes of m × n bits are generated for each block. JPEG2000 performs compression encoding by independently performing entropy encoding with this bit plane as one unit. For example, if the input image is decomposed into the number of blocks N, N × P bit planes are generated, and the input image is converted (encoded) into N × P entropy codes by JPEG2000. Will be.
[0004]
Here, an example in which an input image is decomposed by two-dimensional subband decomposition using JPEG2000 or the like and each subband is divided into code blocks will be described with reference to FIG. FIG. 11A shows an example in which an input image is decomposed into two-dimensional spatial frequency subbands by two-dimensional wavelet transform. FIG. 11B shows an example in which a subband is divided into code blocks of a specific size. As shown in FIG. 11A, the two-dimensional wavelet transform is to apply a one-dimensional wavelet transform to an input image horizontally and vertically, and the input image is transformed into L1L1, L1H1, H1L1, and H1H1 by one decomposition. Decompose. By further performing this decomposition on L1L1, L1L1 is decomposed into L2L2, L2H2, H2L2, and H2H2.
[0005]
By thus repeatedly decomposing the low-frequency region, the two-dimensional wavelet transform realizes hierarchical subband decomposition of the input image. The purpose of performing the sub-band decomposition is to enable a low-resolution image to be reproduced by simply cutting out a part of the encoded single encoded code stream on the receiving side. . Then, the subbands decomposed as shown in FIG. 11A are divided into code blocks of a specific size as shown in FIG. 11B.
[0006]
The input image divided into code blocks in this manner is encoded in units of bit planes, so that the entire image is encoded. Since the code amount of the entire image differs for each image, in order to limit the code amount to a certain code amount or less, it is necessary to selectively discard bit planes so that the distortion amount is minimized as the entire image. Has been done. Performing encoding by selectively discarding the bit plane is called bit plane encoding.
[0007]
Here, the operation of bitplane encoding used in JPEG2000 will be described with reference to FIG. FIG. 12 is a diagram schematically illustrating bit plane encoding in which an input image is divided into code blocks, and bit planes for each code block are selectively discarded to reduce the total code amount to a certain code amount or less. is there.
As shown in FIG. 12, the input image Img is divided into N code blocks B (B (j) (j = 1,..., N)). Each code block B is composed of P bit planes of m × n bits.
[0008]
Here, the bit position of interest in the code block B is set to t (t = 0,..., P−1), and the code amount R when the code block B is coded by discarding the bit planes lower than t. Is defined by equation (1).
[0009]
R (t, j) ≡: r (t, j) +... + R (P−1, j) (1)
[0010]
Note that r (t, j) indicates a code amount when the bit plane t is entropy-coded in the code block B (j) (j = 1,..., N). That is, the code amount R indicates the code amount from the bit plane t to the bit plane (P-1) of the code block B. Since the focused bit position t indicates a reference point at which the bit plane is discarded, it is hereinafter referred to as a TP (Truncation Point).
Also, the amount of distortion D discarded in equation (1) is defined by equation (2).
[0011]
D (t, j) ≡: b (0, j) * 2 ⁰ + ... + b (t-1, j) * 2 ^t-1 … (2)
[0012]
Note that b (t, j) indicates the total number of non-zero bits of the bit plane t in the code block B (j) (j = 1,..., N). That is, the distortion amount D indicates the sum of the distortion amounts from the bit plane 0 to the bit plane (t−1) of the code block B. This distortion amount D is, for example, b (t, j) * 2 ^t Alternatively, the calculation may be performed using the square value.
[0013]
In general, the sub-band (coefficient) decomposed by the sub-band decomposition can take positive or negative values. Therefore, the most significant bit plane (P-1) is set as a bit plane (coefficient sign plane) indicating the sign of the coefficient, and the bit plane .., P-2 are the bit values of the absolute value of the coefficient.
[0014]
For example, when TP is t = 0, no bit plane is discarded, so that D (0, j) = 0 in equation (2). When TP = P−1, all bit planes are discarded, so that R (P−1, j) = 0 in equation (1).
[0015]
Here, FIG. 13 shows a relationship between TP when the number of bit planes of the code block B is 16 (P = 16), and the code amount R (t) and the distortion amount D (t). In FIG. 13, the position t = 15 of the TP of the bit plane is set as the coefficient code plane. Here, assuming that the bit plane TP is t, t bit planes 0,..., T−1 lower than t are discarded. Also, r (s) (0 ≦ s ≦ 15) indicates a code amount when the bit plane s is entropy-coded, and d (s) (0 ≦ s ≦ 15) indicates a case where the bit plane s is discarded. Represents the amount of distortion. This distortion amount is, for example, (total number of non-zero bits of bit plane s) * 2 ^s Or its squared value. R (t) indicates the code amount of the entire code block B (j) (j = 1,..., N) when the position of the TP on the bit plane is t, and D (t) indicates the code block. B (j) indicates the overall distortion amount.
[0016]
Further, FIG. 14 shows a model of the relationship between the position (size) of the TP on the bit plane and the amount of distortion D (t) and the amount of code R (t) for a certain code block B. FIG. 14A shows the relationship between the position (size) of the TP on the bit plane and the distortion amount D (t), and FIG. 14B shows the position (size) of the TP on the bit plane and the code amount R (t). ). As shown in FIG. 14, when TP increases (many bit planes are discarded from the upper bit plane), the distortion amount D (t) increases and the code amount R (t) decreases. Conversely, when the TP becomes smaller (a smaller bit plane is discarded in the lower bit plane), the distortion amount D (t) decreases and the code amount R (t) increases.
Here, for the code block B (j) (j = 1,..., N), a list (TP list) T of t (j) indicating the position of the TP of the code block is defined by Expression (3).
[0017]
T≡ (t (1),..., T (N)) (3)
[0018]
That is, assuming that the position of the TP of the code block B (j) is a bit plane t (j), the total code amount R (T) of the entire N code blocks is given by Expression (4), and the total distortion amount D (T) Can be expressed by equation (5).
[0019]
R (T) = R (t (1), 1) +... + R (t (N), N) (4)
[0020]
D (T) = D (t (1), 1) +... + D (t (N), N) (5)
[0021]
As a result, under the condition that the code amount is limited by JPEG2000 or the like, the solution of the following (Problem A) may be obtained in order to perform truncation of the encoded data and compression encoding of the image.
[0022]
(Issue A)
When the target code amount upper limit is K, a list (TP list) T of TP (Truncation Point) that minimizes D (T) is determined under the condition of R (T) = K (T , R (T) and D (T) refer to equations (3), (4) and (5)).
(End of Question A)
[0023]
Since directly calculating the solution of this (problem A) requires an enormous number of combinational operations, conventionally, in JPEG2000, the solution of the following (problem B) is obtained as a similar problem to (problem A), and a list ( TP list) T (Equation (3)).
[0024]
(Issue B)
Assuming that the target code amount upper limit is K, a combination T and λ that minimizes the evaluation function shown in Expression (6) is obtained under the condition of R (T) = K.
[0025]
L (T, λ) ≡R (T) + λD (T) (6)
[0026]
(For T, R (T) and D (T), refer to equations (3), (4) and (5))
(End of Question B)
[0027]
In order to find a solution to this (problem B), generally, if the evaluation function L (T, λ) is a continuous function, it can be analytically solved using the Lagrange's undetermined multiplier method. It is possible. However, since the evaluation function L (T, λ) is a discrete function, by changing the value of λ according to a plurality of λ values set in advance, the TP for making the evaluation function L (T, λ) smaller is changed. , A list T (t (1),..., T (N)) is sequentially determined.
[0028]
As described above, conventionally, a list T of TPs is determined by an approximate method, bit planes lower than t (1),..., T (N) are discarded for each code block, and each bit plane is discarded. By performing the encoding, the image is compression-encoded under the condition that the code amount is limited.
[0029]
[Problems to be solved by the invention]
However, in the above-described conventional technique, the solution of (Problem A) is the solution of (Problem B), but the solution of (Problem B) is not guaranteed to be the solution of (Problem A) originally desired. That is, the solution of (problem B) is obtained by obtaining a list T (t (1),..., T (N)) of TP (Truncation Point) by an approximate method, and therefore, the evaluation function L (T, λ) (( 6) Refer to the equation).
[0030]
As a method of directly obtaining the solution of (problem A), there is a method of obtaining a solution by a full search. That is, in N code blocks B (j) (j = 1,..., N), R (T) and D (T) with respect to a list T of a certain TP (see equations (4) and (5)) In this method, a list T that satisfies R (T) = K (K is the code amount upper limit) and minimizes D (T) is obtained by sequentially changing the list T.
[0031]
However, in this method, for example, for an image of 8 bits per pixel with 512 horizontal pixels and 512 vertical pixels, encoding is performed by dividing the code block into 64, that is, the image is divided into 64 horizontal pixels and 64 vertical pixels. If the code block is divided into 64 code blocks (N = 64) and the number of bit planes in each code block is 8 (P = 8), the number of possible lists T is P ^N = 6.2 * 10 ⁵⁷ Even if the verification of one list T is performed in 1 nanosecond (ns), to search for the list T that minimizes D (T), 6.2 * 10 ⁴⁸ Second (sec) = 2 * 10 ⁴¹ It takes years and cannot be searched in real time.
[0032]
SUMMARY OF THE INVENTION The present invention has been made in view of the above-described problems, and when an image is compressed and encoded by truncating encoded data under a condition where the code amount is limited, the image is identified. When deciding the bit position at which the bit plane of each block is to be discarded when determining the bit position at which the bit plane of each block is to be discarded, the optimal bit position with the minimum distortion amount is calculated with a much smaller amount of calculation than the full search. It is an object of the present invention to provide a code amount control method, a code amount control device, and a code amount control program that can be obtained.
[0033]
[Means for Solving the Problems]
The present invention has been devised to achieve the above object. First, the code amount control method according to claim 1 divides an image to be encoded into blocks of a specific size, In image coding for generating a variable length code in the order of the divided blocks, a code amount control method of controlling the code amount of each block in order to encode an image within a given code amount upper limit value range. Therefore, the method includes the following steps.
[0034]
That is, a code amount when a block is encoded based on a plurality of quantization levels for quantizing the block and a distortion amount discarded from the block based on the quantization level at that time are calculated, and the distortion Generating a code amount table as a code amount table; a code amount reference value list generating step of generating a code amount reference value list including a plurality of code amount reference values having a maximum code amount upper limit; The quantization level at which the code amount is closest to the code amount reference value in the first block and the distortion amount at that time are set as initial values of an index table and a total distortion amount table in which the code amount reference value is associated with the block. For the work table initial value setting step and for each code amount reference value, the total distortion amount from the first block to the specific block is minimized by referring to the total distortion amount table. A work table setting step of sequentially determining the quantization level and the total amount of distortion at that time, and setting values of an index table and a total amount of distortion table in the second and subsequent blocks, and an index table set in the work table setting step. And a reverse search step of performing a reverse search for a quantization level that minimizes the total distortion amount of all blocks.
[0035]
According to this method, the code amount control method performs the distortion / code amount table generation step, wherein the code amount when the block is encoded based on the plurality of quantization levels and the block amount based on the quantization level at that time. , And the amount of distortion to be discarded is calculated to generate a distortion / code amount table. This quantization level indicates a level for reducing the amount of code at the time of encoding.
[0036]
In the code amount control method, the code amount reference value list including a plurality of code amount reference values having the maximum code amount upper limit value is generated in the code amount reference value list generation step. This code amount reference value is used to solve the problem of finding the optimum quantization level at the code amount upper limit value, to solve the small problem, and to solve the problem to be obtained by sequentially expanding the scale of the problem from the result. In addition, the code amount upper limit value is subdivided so as to be a small problem value. For example, when encoding an input image of HDTV (High Definition Television), the number of code amount reference values is set to about 100 to 200.
[0037]
Further, in the code amount control method, the work table initial value setting step sets the quantization level at which the code amount is closest to the code amount reference value in the first block and the distortion amount at that time to the code amount reference value. Blocks are set as initial values of an index table and a total distortion amount table, and a work table setting step refers to the total distortion amount table for each code amount reference value, and refers to the total distortion amount table from the first block to the specific block. The quantization level that minimizes the total distortion amount and the total distortion amount at that time are sequentially determined, and the values of the index table and the total distortion amount table in the second and subsequent blocks are set. Thus, the index table and the total distortion amount table initialized in the work table initial value setting step are recursively set in the order of blocks in the work table initial value setting step.
[0038]
Then, in the code amount control method, in a reverse search step, a quantization level that minimizes the total distortion amount of all blocks is reversely searched for from the index table set in the work table setting step, and is obtained. In this reverse search step, since the index table has already been set in the work table setting step, by referring to this index table, the quantization level that minimizes the total distortion amount is searched in order from the last block, Get the quantization levels for all blocks.
[0039]
In the code amount control method according to the present invention, an image to be encoded is divided into blocks B (j) (j = 1,..., N) of a specific size, and each block is variable. In the image encoding for generating a long code, a code amount control method for controlling the code amount of each block in order to encode an image within a given code amount upper limit K, comprising the following steps: Was included.
[0040]
That is, the code amount R (t, j) when the block B (j) is encoded based on a plurality of quantization levels t for quantizing the block, and the block B (j) based on the quantization level t A distortion / code amount table generation step of calculating the amount of distortion D (t, j) to be discarded and generating the same as a distortion / code amount table, and S (K) codes with the maximum code amount upper limit K as the maximum value A code amount reference value list generating step of generating a code amount reference value list composed of the amount reference values k (i) (i = 1,..., S (K) = K); An index table that associates a quantization level t with a close code amount R (t, 1) and a distortion amount D (t, 1) at that time with a code amount reference value k (i) and a block B (j). Idx (i, j) and the work to be set as initial values of the total distortion amount table F (i, j) Including a table initial value setting step.
[0041]
Further, a code amount is defined as r, a mapping that returns an index i of the code amount reference value k (i) closest to the r is defined as ψ (r), and a total distortion amount f (i, j; t) is defined as D (t, t). j) + F (ψ (k (i) −R (t, j)), j−1), and the quantization level t that minimizes the total distortion amount f (i, j; t), and A work table setting step of setting the total distortion amount f (i, j; t) in the index table Idx (i, j) and the total distortion amount table F (i, j), and the index of the code amount reference value k (i). S (K) is given to i as an initial value and N is given to an index j of a block B (j). A quantization level t (j) assigned to the block B (j) is set as an index table Idx (i, j). ψ (k (i) -R (t (j), j)) is calculated, and the calculation result is again calculated for i. And an inverse search step of performing an inverse search of the list (t (1),..., T (N)) of the quantization levels t (j) from the index table Idx (i, j) by sequentially subtracting j. , Is included.
[0042]
According to this method, the code amount control method uses the code amount R () when the block B (j) is encoded based on the plurality of quantization levels t for quantizing the block in the distortion / code amount table generation step. t, j) and the amount of distortion D (t, j) discarded from the block B (j) based on the quantization level t, and are generated as a distortion / code amount table.
[0043]
In the code amount control method, a code amount reference value list is generated in a code amount reference value list generation step. This code amount reference value list is composed of S (K) code amount reference values k (i) (i = 1,..., S (K) = K) having the maximum code amount upper limit K as the maximum value.
[0044]
Further, in the code amount control method, the index table Idx (i, j) and the total distortion amount table F (i) in which the code amount reference value k (i) and the block B (j) are associated in the work table initial value setting step. , J) are initialized. The initial values used in this initialization are the quantization level t at which the code amount R (t, 1) is closest to the code amount reference value k (i) and the distortion amount D (t, 1) at that time. Then, in the work table setting step, values are set in the index table Idx (i, j) and the total distortion amount table F (i, j). The values set in the index table Idx (i, j) and the total distortion amount table F (i, j) return the index i of the code amount reference value k (i) closest to the map ψ (r). A total distortion amount f (i, j; t) is defined as D (t, j) + F (ψ (k (i) −R (t, j)), j−1). A quantization level t that minimizes (i, j; t) and a total distortion amount f (i, j; t) at that time are set. The mapping ψ (r) is for calculating the difference between the code amount r and the code amount reference value k (i), and returning the index i that minimizes the difference amount.
[0045]
Then, in the code amount control method, a list (t (1),..., T (N)) of the quantization levels t (j) is reversely searched from the index table Idx (i, j) by the reverse search step. In the reverse search, S (K) is given to index i of code amount reference value k (i), N is given to index j of block B (j) as an initial value, and quantization level t assigned to block B (j) is given. (J) is used as an index table Idx (i, j), a mapping ψ (k (i) −R (t (j), j)) is calculated, the calculation result is again set to the value of i, and j is sequentially subtracted. Do it by doing.
[0046]
Further, the code amount control method according to claim 3 is the code amount control method according to claim 2, wherein the quantization level t is a value indicating the position of the bit plane of each block B (j). , Each block B (j) is quantized based on the position of the bit plane.
[0047]
According to this method, the code amount control method sets the quantization level t to a value indicating the position of the bit plane of each block B (j), and discards the bit plane based on this value, thereby making each block B (j) discard. B (j) is quantized. This reduces the amount of code in encoding.
[0048]
The code amount control device according to claim 4 divides an image to be encoded into blocks of a specific size, and generates a variable-length code in the order of the divided blocks. A code amount control device that controls the code amount of each block in order to encode an image within a given code amount upper limit range, and has the following configuration.
[0049]
That is, the amount of code when the block is coded based on a plurality of quantization levels for quantizing the block, and the amount of distortion discarded from the block based on the quantization level are calculated. A distortion / code amount table generating means for generating, a distortion / code amount table holding means for holding the distortion / code amount table generated by the distortion / code amount table generating means, and a plurality of codes having a maximum code amount upper limit value Code amount reference value list generating means for generating a code amount reference value list composed of the code amount reference values, and a code amount reference value list holding the code amount reference value list generated by the code amount reference value list generating means Means, the quantization level of the specific block in which the code amount from the first block to the specific block is the code amount closest to the code amount reference value, and the total distortion amount at that time, A work table generating unit that generates an index table and a total distortion amount table in which the signal amount reference value and the block are associated with each other; a work table holding unit that holds the index table and the total distortion amount table generated by the work table generating unit; A reverse search means for performing a reverse search for a quantization level which minimizes the total distortion amount of all blocks from the index table held in the work table holding means.
[0050]
According to this configuration, the code amount control device discards the code amount when the block is coded based on the plurality of quantization levels by the distortion / code amount table generation unit, and discards the block based on the quantization level. The amount of distortion is calculated as a distortion / code amount table, and stored in the distortion / code amount table holding means. This quantization level indicates a level for reducing the amount of code at the time of encoding.
[0051]
The code amount control device generates a code amount reference value list including a plurality of code amount reference values having the maximum code amount upper limit value by the code amount reference value list generation unit, and stores the generated code amount reference value list in the code amount reference value list holding unit. Hold. This code amount reference value is used to solve the problem of finding the optimum quantization level at the code amount upper limit value, to solve the small problem, and to solve the problem to be obtained by sequentially expanding the scale of the problem from the result. In addition, the code amount upper limit value is subdivided in order to make it a small problem value. For example, when encoding an input image of HDTV (High Definition Television), the number of code amount reference values is set to about 100 to 200.
[0052]
Further, the code amount control device uses the work table generating means to set the quantization level of the specific block in which the code amount from the first block to the specific block is the code amount closest to the code amount reference value and the distortion at that time. The total amount is generated as an index table and a total distortion amount table in which a code amount reference value and a block are associated with each other, and is held in a work table holding unit.
[0053]
Then, the code amount control device reversely searches for a quantization level that minimizes the total distortion amount of all blocks from the index table held in the work table holding means by the reverse search means. In this reverse search means, the index table is already held in the work table holding means by the work table generation means, and by referring to this index table, the quantization levels that minimize the total amount of distortion are sequentially determined from the last block. To obtain the quantization levels of all the blocks.
[0054]
Further, the code amount control program according to claim 5 divides an image to be encoded into blocks of a specific size, and generates a variable length code in the order of the divided blocks. In order to control the code amount of each block when encoding an image within the range of the given code amount upper limit, the computer is configured to function by the following means.
[0055]
That is, the amount of code when the block is coded based on a plurality of quantization levels for quantizing the block, and the amount of distortion discarded from the block based on the quantization level are calculated. A distortion / code amount table generating unit that generates and stores the distortion / code amount table in the distortion / code amount table holding unit, generates a code amount reference value list including a plurality of code amount reference values having the maximum code amount upper limit value, and generates a code amount reference value. Code amount reference value list generation means held in the value list holding means, the quantization level at which the code amount from the first block to the specific block is the code amount closest to the code amount reference value, and the total distortion amount at that time Table generating means for generating an index table and a total distortion amount table in which a code amount reference value and a block are associated with each other, and storing the generated total table in a work table holding means. From the index table held by lifting means and minimize distortion total amount of all blocks reverse search means for obtaining by reverse search the quantization level, and.
[0056]
According to this configuration, the code amount control program is discarded by the distortion / code amount table generation unit from the block based on the code amount when the block is encoded based on the plurality of quantization levels and the quantization level. The amount of distortion is calculated as a distortion / code amount table, and stored in the distortion / code amount table holding means. This quantization level indicates a level for reducing the amount of code at the time of encoding.
[0057]
The code amount control program generates a code amount reference value list including a plurality of code amount reference values having the maximum code amount upper limit value by the code amount reference value list generation unit, and stores the generated code amount reference value list in the code amount reference value list holding unit. Hold. This code amount reference value is used to solve the problem of finding the optimum quantization level at the code amount upper limit value, to solve the small problem, and to solve the problem to be obtained by sequentially expanding the scale of the problem from the result. In addition, the code amount upper limit value is subdivided in order to make it a small problem value. For example, when encoding an input image of HDTV (High Definition Television), the number of code amount reference values is set to about 100 to 200.
[0058]
Further, the code amount control program uses the work table generating means to set the quantization level at which the code amount from the first block to the specific block is the code amount closest to the code amount reference value and the total distortion amount at that time. An index table and a total distortion amount table in which a code amount reference value and a block are associated with each other are generated and stored in a work table storage unit.
[0059]
Then, the code amount control program reversely searches for a quantization level that minimizes the total distortion amount of all blocks from the index table held in the work table holding means by the reverse search means. In this reverse search means, the index table is already held in the work table holding means by the work table generation means, and by referring to this index table, the quantization levels that minimize the total amount of distortion are sequentially determined from the last block. To obtain the quantization levels of all the blocks.
[0060]
BEST MODE FOR CARRYING OUT THE INVENTION
Hereinafter, embodiments of the present invention will be described with reference to the drawings.
FIG. 1 is a block diagram showing a configuration of an encoding device including a code amount control device according to the present invention. An encoding device 1 shown in FIG. 1 includes a subband decomposition unit 2, a code amount control device 3, and an entropy code generation unit 4, and a subband code using a wavelet transform represented by the image encoding standard JPEG2000. In this way, the input image is compression-coded within the given code amount upper limit value and is converted into an entropy code.
[0061]
The sub-band decomposition unit 2 decomposes the input image by two-dimensional sub-band decomposition using wavelet transform. As shown in FIG. 11A, the sub-band decomposition unit 2 performs one-dimensional wavelet transform on an input image horizontally and vertically, and performs L1L1, L1H1, H1L1, H1L1, Decomposes into H1H1. By further performing this decomposition on L1L1, L1L1 is decomposed into L2L2, L2H2, H2L2, and H2H2. As described above, the subband decomposition unit 2 decomposes the input image into hierarchical subbands by repeatedly decomposing the low frequency region of the input image.
[0062]
The code amount control device 3 divides the sub-band decomposed by the sub-band decomposing unit 2 into blocks (code blocks) of a specific size, and changes the quantization level (quantization level) for each of the blocks. When the data is truncated by performing the quantization, the quantization level for each block that minimizes the coding distortion is selected. Here, the quantization level indicates the position of the bit plane used in the bit plane coding described in FIG. The configuration of the code amount control device 3 will be described later.
[0063]
The entropy code generation means 4 performs entropy coding such as adaptive arithmetic coding on the data reduced by the quantization level of each block in the code amount control device 3. The entropy code generation means 4 generates an entropy code which is encoded data.
[0064]
(Configuration of code amount control device)
Next, the configuration of the code amount control device according to the present invention will be described.
As shown in FIG. 1, the code amount control device 3 includes a code block dividing unit 10, a distortion / code amount table generating unit 11, a code amount reference value list generating unit 12, a work table generating unit 13, a reverse search, Means 14 and a bit truncation means 15.
[0065]
The code block dividing means 10 divides the input image decomposed into sub-bands by the sub-band decomposing means 2 into blocks (code blocks) of a specific size (see FIG. 11B). Here, the code block dividing means 10 expresses the pixels of each sub-band by P bits, and furthermore, each sub-band is spatially N code blocks B (j) (j = (n = n) of m pixels vertically and n pixels horizontally). 1,..., N). Thereby, each code block is generated as P bit planes of m × n bits. This code block is output to the distortion / code amount table generation unit 11 and the bit truncation unit 15.
[0066]
The distortion / code amount table generation unit 11 changes the quantization level of the code block divided by the code block division unit 10 to perform the entropy encoding, and the distortion amount discarded by the encoding. To generate a distortion / code amount table in which each code block is associated with a quantization level. Note that this entropy coding is the same coding as the entropy code generation means 4. Here, as the quantization level, a position (TP: Truncation Point) serving as a reference for discarding a bit plane in the bit plane coding described with reference to FIG. 12 is used.
[0067]
In the distortion / code amount table generation means 11, the TP of the code block B (j) is set to t (t = 0,..., P-1) as a distortion / code amount table, and bit planes lower than t are discarded. The code amount when the code block is coded by using Equation (1) is defined as a code amount table (R (t, j)), and the discarded amount of distortion in Equation (1) is calculated by using the distortion amount table ( D (t, j)). That is, as shown in FIG. 2, the distortion amount table (see FIG. 2A) and the code amount table (see FIG. 2B) include an index t indicating the position of the TP in the P bit planes. , An index j indicating the code block B (j). The distortion amount table and the code amount table correspond to a distortion / code amount table holding unit described in claims, and are stored in a storage unit such as a memory.
[0068]
The code amount reference value list generation means 12 generates a sequence having the maximum value of the given code amount upper limit as a code amount reference value list (code amount reference value list Kset). Assuming that the given code amount upper limit value is K, the code amount reference value list generation unit 12 generates an S (K) type code amount reference value list Kset with K being the maximum value as shown in FIG. k (1),..., k (S (K)) are generated as one-dimensional array data. For example, when encoding an input image of HDTV (High Definition Television), the number of sequences in the code amount reference value list Kset is set to about 100 to 200. The code amount reference value list Kset corresponds to a code amount reference value list holding unit described in claims, and is held in a storage unit such as a memory.
[0069]
The work table generation unit 13 refers to the distortion amount table D, the code amount table R generated by the distortion / code amount table generation unit 11, and the code amount reference value list Kset generated by the code amount reference value list generation unit 12. For each code amount reference value k (i) (i = 0,..., S (K)), quantization is performed so that the total distortion amount of j code blocks B (1),. The total amount of distortion when the level list is set is defined as a total distortion amount table F (i, j), and an index indicating the quantization level (TP) at that time is generated as an index table Idx (i, j). The procedure for generating the total distortion amount table F and the index table Idx will be described later. The distortion total amount table F and the index table Idx correspond to a work table holding unit described in claims, and are held in a storage unit such as a memory.
[0070]
The inverse search means 14 generates the code amount table R generated by the distortion / code amount table generation means 11, the code amount reference value list Kset generated by the code amount reference value list generation means 12, and the work table generation means 13. With reference to the index table Idx, the position (t (1),..., T (N)) of the TP for each of the N code blocks B (j) (j = 1,. To decide.
[0071]
The reverse search means 14 calculates the value of Idx (S (K), N) already obtained from the index table Idx generated by the work table generation means 13, that is, the number of code blocks is N. , T (1) are sequentially searched from the index table Idx by using the position of the TP (t (N)) of the code block B (N) as an initial value. is there. This reverse search procedure will be described later.
[0072]
The bit truncation unit 15 converts the code block B (j) (j = 1,..., N) divided by the code block division unit 10 into the TP list T (t (1),. , T (N)), the bit planes lower than the position of TP in the TP list T are discarded from each code block B (j), and the remaining bit planes are output to the entropy code generation means 4. .
As a result, the entropy code encoded by the entropy code generation unit 4 becomes encoded data obtained by compressing and encoding the input image with a minimum distortion amount within a given code amount upper limit value.
[0073]
As described above, the configuration of the code amount control device 3 has been described based on one embodiment. However, the code amount control device 3 can also realize each unit in a computer as each function program, and combine each function program. Then, it can be operated as a code amount control program.
[0074]
(Operation of code amount control device)
Next, the operation of the code amount control device 3 will be described.
[Operation concept]
First, the concept of the code amount control operation performed by the code amount control device 3 will be described. The code amount control device 3 calculates (Solution A) with N code blocks B (j) (j = 1,..., N) as shown in FIG. The solution of (problem A) is finally obtained by decomposing into N partial objects and sequentially solving the problems in the partial objects.
[0075]
The code amount reference values k (i) (i = 1,..., S (K)) generated by the code amount reference value list generation unit 12 are used for the N objects (1 to N) shown in FIG. The problem of optimizing the allocation of the code amount of the target j (j = 1,..., N), which is the upper limit, is represented by A (i, j). Formulated as (sub-problem A).
[0076]
(Sub-problem A)
For the j code blocks B (1), B (2),..., B (j), when the target code amount upper limit is k (i), R (T (j)) = k (i) ), A list (TP list) T (j) of TPs (Truncation Points) that minimizes D (T (j)) is determined (R and D are expressed by equations (4) and (5)). See formula).
Here, T (j) = ((t (1),..., T (j))), and TPs assigned to j code blocks B (1), B (2),. As a list T (j).
(End of sub-problem A)
[0077]
Of the S (K) × N subproblems formulated in this way, (subproblem A) A (S (K), N) corresponds to the originally desired (problem A).
The code amount control device 3 refers to the already solved small-scale (sub-problem A) solution while sequentially increasing i and j from the smallest (sub-problem A) A (1, 1). By doing so, the problem is solved successively, and finally (sub-problem A) A (S (K), N) desired to be obtained, that is, (problem A) is solved.
[0078]
Next, the operation of the code amount control device 3 for sequentially solving (sub-problem A) will be described with reference to FIG. FIG. 5 is a flowchart illustrating a schematic operation of the code amount control device 3. The operations of the code block dividing means 10 and the bit truncating means 15 are the same as those in the conventional art, and thus the description is omitted here.
[0079]
[Step S1: Distortion / code amount table generation]
First, the code amount control device 3 divides each of the code blocks B (j) (j = 1,..., N) divided by the block dividing means 10 as P bit planes of m × n bits into distortion / code amounts. The table generation means 11 generates a code amount subjected to entropy coding for each quantization level (TP) value (t = 0,..., P-1) as a code amount table R (t, j). The distortion amount discarded by encoding is generated as a distortion amount table D (t, j).
[0080]
[Step S2: Code amount reference value list generation]
When the code amount reference value list generating means 12 sets the given code amount upper limit value to K, the code amount control device 3 lists S (K) types of code amount reference values in which K is the maximum value. (Code amount reference value list Kset) is generated as Kset≡ (k (1),..., K (S (K)) = K).
[0081]
Note that the code amount reference value list Kset only needs to be k (i) <k (i + 1) and k (S (K)) = K. For example, let the positive number ε be K / (S (K) −1), and let k (i) be a sequence that increases linearly as shown in equation (7).
[0082]
k (i) = (i−1) * ε (7)
[0083]
Further, the code amount reference value list Kset may be a non-linearly increasing numerical sequence that divides the vicinity of a certain code amount reference value into small pieces.
[0084]
[Step S3: Work table initial value setting]
Then, the code amount control device 3 causes the work table generating means 13 to sequentially and recursively solve the (sub-problem A), an index table Idx of S (K) rows and N columns, and a total distortion table F Set the initial value to.
[0085]
Here, Idx (i, j) and F (i, j) (i = 1,..., S (K), j = 1,..., N) are stored in the index table Idx and the i-th row j of the total distortion amount table F. As a column element, F (i, j) is a list T (j) of optimal TPs for j code blocks B (1),..., B (j) with a code amount reference value k (i). When this is set, that is, (sub-problem A), it is a place to store the total distortion amount D (T (j)) when solving A (i, j). Idx (i, j) is a place for storing the value of t (j) indicating the TP of the code block B (j) when solving (sub-problem A) A (i, j).
[0086]
In this step S3, focusing on the code block B (1), the smallest (sub-problem A) A (i, 1) (i = 1,..., S (K)) is solved, and the index tables Idx and The element in the first column of the total distortion amount table F is determined. This (sub-problem A) A (i, 1) (i = 1,..., S (K)) is a problem of allocating a code amount to one block B (1) at a certain code amount reference value k (i). In T (1), which is a list of TPs to be obtained, there is only one element t (1). Therefore, (sub-problem A) the solution of A (i, 1), that is, t (1) which is the optimal TP for code block B (1) under the code amount constraint of the code amount reference value k (i) is , Code amount total R (T) = k (i), that is, R (t (1), 1) = k (i). In this case, t (1) for which R (t (1), 1) = k (i) is not always true does not always exist, and here, the code amount R “closest” to k (i) is used. The value of t (1), that is, t (1), is determined, and the value of t (1) is stored in the index table Idx (i, 1).
[0087]
F (i, 1) of the total distortion amount table F includes (sub-problem A) the distortion amount in A (i, 1), that is, the code block B (1) in Idx (i, 1) of the index table Idx. ), The value of the distortion amount D (Idx (i, 1), 1) is read from the distortion amount table D and stored.
[0088]
Here, “closest” is defined. For example, when the code amount is r and the code amount reference value is k (i) (i = 1,..., S (K)), the closeness measure is | k (i) −r | (k (i). Assuming that (the absolute value of the difference from r), k (i) monotonically increases (k (i−1) <k (i)) with respect to i as shown in equation (7), and k (i) -1) <r <k (i), and r is located exactly between k (i-1) and k (i), that is, rk (i-1) = k (i) If -r, the value “closest” to r is k (i−1) and k (i), and the code amount reference value “closest” to a certain code amount r cannot be specified. become.
[0089]
Therefore, here, | r−k (i) | ≦ | r−k (i ′) | is established for all i ′s where k (i) ≦ r and i ′ ≠ i, so that a certain code It is assumed that i of the code amount reference value k (i) “closest” to the amount r is specified. Note that there are various other methods for specifying i. For example, | r−k (i) | ≦ | r for all i ′ where k (i) ≧ r and i ′ ≠ i −k (i ′) |.
[0090]
The operation of the work table initial value setting (step S3) is shown in the flowchart of FIG. As shown in FIG. 6, in the work table initial value setting (step S3), first, a variable i is initialized to 1 (step S30). Then, the value of TP for the code block B (1) having the code amount R (t, 1) "closest" to the code amount reference value k (i) is set to t (step S31). In step S31, the argument {condition | parameter list} represents an operation of extracting a parameter satisfying the "condition" from the "parameter list". In addition, the “≒” symbol in the “condition” is a symbol representing “closest”.
[0091]
Then, the value t of the TP acquired in step S31 is stored in the index table Idx (i, 1) (step S32), and the distortion amount D (t, 1), that is, D (Idx (i, 1), 1). Is read and stored in F (i, 1) of the total distortion amount table F (step S33). Then, it is determined whether or not the variable i is the number of code amount reference values (S (K)) (step S34). If the variable i is equal (Yes), the process is terminated, and if different (No), the value of the variable i is set to 1 It adds (step S35), returns to step S31, and continues processing.
[0092]
By the above operation, in the code amount reference value k (i) (i = 1,..., S (K)), Idx (i, 1) of the index table Idx includes the code block B (1). The value of TP that gives the “closest” code amount to k (i) is stored. Also, F (i, 1) of the total distortion amount table F indicates the distortion amount of the code block B (1) in the TP that gives the code amount “closest” to k (i) with respect to the code block B (1). Is stored.
Thereby, the S (K) values in the first column in the index table Idx and the total distortion amount table F are filled, and (sub-problem A) A (i, 1) (i = 1,..., S (K) ) Has been solved.
[0093]
[Step S4: Work table setting]
Next, the code amount control device 3 uses the result of the (sub-problem A) A (i, 1) solved in step S3 by the work table generating means 13 as an initial value, and sets the code blocks B (i),. With reference to (sub-problem A) A (i, j-1) related to (j-1), (sub-problem A) A (i, j) is solved. That is, the value of the j-th column is determined recursively by referring to the S (K) values of the (j-1) -th column already determined in the index table Idx and the total distortion amount table F. .
[0094]
Here, the operation of setting the work table (step S4) will be described with reference to the flowchart of FIG.
As shown in FIG. 7, in the work table setting (step S4), first, a variable j which is an index of the code block B (j) is initialized to 2 (step S40), and an index of the code amount reference value k (i) is set. Is initialized to 1 (step S41). The reason why the variable j is initialized to 2 is that the index table Idx in the code block B (1) has already been set in the work table initial value setting (step S3).
[0095]
(Sub-problem A) In A (i, j), when the last element t (j) of T (j) which is a list of TPs is t (the TP of the code block B (j) is t) , B (j) are calculated (step S42).
Here, first, a mapping の from the code amount r to the index of the element of the code amount reference value list Kset is defined by Expression (8).
[0096]

[0097]
In the expression (8), the argument {condition | parameter list} represents an operation of extracting a parameter satisfying the “condition” from the “parameter list”, and the “≒” symbol in the “condition” is a symbol representing “closest”. And
Formula (9) is used to calculate the total distortion amount f (i, j; t) of the code blocks B (1),..., B (j) in step S42.
[0098]

[0099]
In this equation (9), (k (i) -R (t, j)) is (sub-problem A) from the code amount reference value k (i) assigned to A (i, j), The amount obtained by subtracting the code amount R (t, j) of (j) is referred to as a residual code amount here. Ψ (k (i) −R (t, j)) is a code “closest” to the residual code amount (k (i) −R (t, j)) as shown in Expression (8). The index of the element of the quantity reference value list Kset is shown. Therefore, F (ψ (k (i) −R (t, j)), j−1) of the distortion total amount table F is (sub-problem A) A (ψ (k (i) −R (t, j)) , J−1) are solved.
[0100]
Equation (9) shows that the (sub-problem A) A (ψ (k (i) −R (t, j), j−1)) is solved by adding the total distortion to the code block B (j). This is the sum of the distortion amounts D (t, j). (Sub-problem A) In A (i, j), when the last element t (j) of T (j), which is a list of TPs, is t .., B (j).
As described above, the value t of the TP at which the total amount of distortion calculated by the equation (9) is minimized is obtained using the equation (10) (step S43).
[0101]

[0102]
The expression (10) indicates that the value of t that takes the minimum value (minimize) when the total distortion amount f (i, j; t) changes according to the parameter t is extracted. Here, there are at most P (0,..., P−1) elements of the total distortion amount f (i, j; t), and the minimum value among them is (sub-problem A) solving A (i, j). It is the total amount of distortion when there is. Therefore, if the value t of TP when the total amount of distortion becomes the minimum value is obtained, the value of t is (sub-problem A) in the list T (j) of TP which is the solution of A (i, j). It becomes the last element t (j).
[0103]
Then, the value t of the TP obtained in step S43 is stored in the index table Idx (i, j) (step S44), and the total distortion amount f (i, j; t) at the value of t at that time is calculated as the total distortion amount. It is stored in F (i, j) of the table F (step S45). Then, it is determined whether or not the variable i is the number of code amount reference values (S (K)) (step S46). If different (No), the value of the variable i is incremented by 1 (step S47), and the process proceeds to step S42. Return and continue processing.
[0104]
If the variable i is equal to the number of code amount reference values (S (K)) (Yes), it is determined whether the variable j is the number of code blocks (N) (Step S48), and if equal, (Yes). Ends the processing, and if different (No), adds 1 to the value of the variable j (step S49), returns to step S41, and continues the processing.
[0105]
By the above operation, the values of the code blocks B (j) (j = 2,..., N) in the second and subsequent columns in the index table Idx and the total distortion amount table F are filled in order by S (K), and Problem A) When solving A (i, j), Idx (i, j) which is the value of the index table Idx and F (i, j) which is the value of the total distortion amount table F are determined. FIG. 9 shows the order in which the elements of the index table Idx and the total distortion amount table F are sequentially determined. FIG. 9A schematically shows the order in which the index table Idx is determined, and FIG. 9B schematically shows the order in which the total distortion table F is determined. In both the index table Idx and the total distortion amount table F, the indexes i = 1 to i = S (K) of the code amount reference value k (i) of the first column (code block B (1)) are set as initial values. After that, the values from the code block B (2) in the second column to the code block B (N) are determined in order.
[0106]
In this way, by recursively solving (sub-problem A), (sub-problem A) F (S (K), N) which is the value of the total distortion amount table F in A (S (K), N) ) And Idx (S (K), N) which is the value of the index table Idx.
[0107]
[Step S5: reverse search]
Next, the code amount control device 3 causes the work table generating means 13 to output Idx (i, j) (i = 1,...) Which is a table of S (K) rows and N columns of the index table Idx set in step S4. Using S (K), j = 1,..., N), Idx (S (K), N), that is, the value t of the TP in the code block B (N) when the number of code blocks is N Using (N) as an initial value, t (N−1),..., T (1) are sequentially determined by an inverse search.
[0108]
Here, the operation of the reverse search (step S5) will be described with reference to the flowchart of FIG.
As shown in FIG. 8, in the reverse search (step S5), first, a variable i which is an index of the code amount reference value k (i) is initialized to the number of code amount reference values (S (K)) (step S50). ), A variable j, which is an index of the code block B (j), is initialized to the number (N) of code blocks (step S51).
[0109]
Then, Idx (i, j), which is the value of the TP allocated to the code block B (j), is obtained from the index table Idx, and is set to t (j) (step S52). In this step S52, assuming that a certain Idx (i, j) is determined, the code amount reference value k (i) is assigned to the code blocks B (1),..., B (j) (sub-problem A) A ( i, j) has already been solved, and Idx (i, j) becomes t (j) which is the value of the TP allocated to the last code block B (j).
[0110]
As a result, the amount obtained by subtracting the code amount R (t, j) to be allocated to the code block B (j) from the code amount reference value k (i), that is, the remaining code amount k (i) -R (t (j)) , J) are the code amounts to be assigned to the code blocks B (1),..., B (j-1). Therefore, the index of the code amount reference value “closest” to the residual code amount k (i) −R (t (j), j) of the code block B (j) is ψ (k (i) −R (t ( j) and j)), and newly assigned to a variable i (step S53).
[0111]
Then, 1 is subtracted from the variable j which is the index of the code block B (j) (step S54), and it is determined whether or not the variable j has become 0 (step S55). Here, when the variable j has become 0 (Yes), the process is terminated, and when the variable j is not 0 (No), the process returns to step S52 to continue the process.
[0112]
The above operation will be further described with reference to FIG. FIG. 10 is a diagram schematically showing the reverse search operation in the reverse search means 14. The value of the index table Idx, Idx (S (K), N), that is, the value of the TP (t (N)) of the code block B (N) has already been obtained in step S4 (written to A1). There). Next, the remaining code amount (K−R) obtained by subtracting the code amount (R (t (N), N)) when the code block B (N) is encoded with t (N) from the code amount upper limit K. Let i be the index of the element of the code amount reference value list Kset “closest” to (t (N), N)). That is, an index i that satisfies KR (t (N), N) ≒ k (i) is obtained. Then, in the index table Idx, the value (Idx (i, N-1)) of the code block B (N-1) corresponding to the index i (written in A2) is stored in the code block B (N-1). (T (N-1)).
[0113]
Further, from the code amount reference value k (i), the code amount (R (t (N-1), N-1)) when the code block B (N-1) is encoded at t (N-1). Let i ′ be the index of the element of the code amount reference value list Kset that is “closest” to the residual code amount (KR (t (N−1), N−1)) obtained by subtracting. That is, an index i ′ that satisfies KR (t (N−1), N−1) ≒ k (i ′) is obtained. Then, in the index table Idx, the value (Idx (i ', N-2)) of the code block B (N-2) corresponding to the index i' (written in A3) is stored in the code block B (N- The value of TP in 2) is (t (N−2)). In this way, the value of TP is sequentially searched backward until t (1).
[0114]
By the above operation, the list T = (t (1),..., T (N)) of the TPs of the N code blocks B (j) refers to the index table Idx that has already been obtained (reverse search). , And t (N),..., T (1).
By the above operations of steps S1 to S5, the code block B (j) (j = 1, j) that minimizes the distortion sum D (T) (see equation (5)) under the given code amount upper limit K. , N), a list (TP list) T of TPs (Truncation Points) can be sequentially determined.
[0115]
As is clear from the operations in steps S1 to S5, according to this method, (problem A) can be solved by an operation of the order of S (K) * N * P. For example, if the number of code amount reference values is 100 (S (K) = 100), the number of code blocks is 64 (N = 64), and the number of bit planes in each code block is 16 (P = 16), 51200 The operation can be performed in a single calculation step. For example, even if the calculation time is 100 nanoseconds (ns) per step, the present procedure can be operated at a frame rate of 1/30 second. As a result, code amount control by software can be realized.
[0116]
In the present method, the quantization level has been described as a TP (Truncation Point) indicating the position of a bit plane. However, this quantization level is not limited to the bit plane coding used in subband coding. In addition, even in an encoding method in which the code amount and the distortion amount are parameter-controlled in block units, it can be realized by regarding the parameters as quantization levels. For example, in DCT (Discrete Cosine Transform) -based block coding, a quantization level may be determined for each block.
[0117]
Further, in JPEG2000, one bit plane is classified into three types of paths and entropy-encoded, but in this method, it is possible to treat each path as a separate bit plane and treat it in the same manner as the above-described operation. It is possible.
[0118]
【The invention's effect】
As described above, the code amount control method, the code amount control device, and the code amount control program according to the present invention have the following excellent effects.
[0119]
According to the first, second, fourth, or fifth aspect of the present invention, the amount of information is discarded (quantized) to perform compression encoding within a predetermined code amount upper limit. In image coding, a code for minimizing the amount of distortion can be determined by a short-term operation using a table search and four arithmetic operations.
As a result, even a general-purpose CPU can compress and encode a moving image in real time with high image quality, and can realize an encoding device at low cost.
[0120]
According to the third aspect of the present invention, quantization is performed by, for example, bit plane coding by setting the level at which quantization is performed (quantization level) as a reference position at which a bit plane in bit plane coding is discarded. With JPEG2000 or the like, it is possible to divide an image into blocks and control the position of discarding the bit plane of each block for each block.
[Brief description of the drawings]
FIG. 1 is a block diagram showing an overall configuration of a code amount control device according to an embodiment of the present invention.
FIG. 2 is an explanatory diagram for explaining the structures of a distortion amount table and a code amount table according to the embodiment of the present invention.
FIG. 3 is an explanatory diagram for explaining a structure of a code amount reference value list according to the embodiment of the present invention.
FIG. 4 is an explanatory diagram illustrating an operation of disassembling a code block into a plurality of objects in order to obtain a solution to (problem A) according to the embodiment of the present invention.
FIG. 5 is a flowchart schematically showing a code amount control operation according to the embodiment of the present invention.
FIG. 6 is a flowchart showing an operation of initial setting of a work table according to the embodiment of the present invention.
FIG. 7 is a flowchart showing an operation of setting a work table according to the embodiment of the present invention.
FIG. 8 is a flowchart showing an operation of a reverse search according to the embodiment of the present invention.
FIG. 9 is a schematic diagram visually illustrating an order of setting an index table and a total distortion amount table according to the embodiment of the present invention.
FIG. 10 is a schematic diagram schematically showing a reverse search operation according to the embodiment of the present invention.
FIG. 11 is an explanatory diagram for explaining subband decomposition and code block division in subband encoding.
FIG. 12 is an explanatory diagram for describing an outline of an operation of bit plane encoding.
FIG. 13 is an explanatory diagram for explaining a relationship between a truncation point and a code amount and a distortion amount in bit plane coding.
FIG. 14 is an explanatory diagram for explaining the relationship between the position (size) of a bit plane and the amount of distortion and the amount of code.
[Explanation of symbols]
1 ... Encoding device
2 ... Subband decomposition means
3. Code amount control device
4 ... Entropy code generation means
10 code block dividing means
11... Distortion / code amount table generating means
12 Code amount reference value list generation means
13 Work table generating means
14 Reverse search means
15 Bit truncation means
D: Distortion amount table (distortion / code amount table holding means)
R code amount table (distortion / code amount table holding means)
Kset: code amount reference value list (code amount reference value list holding unit)
Idx Index table (work table holding means)
F ...... Total distortion amount table (work table holding means)

Claims (5)

In an image encoding in which an image to be encoded is divided into blocks of a specific size and a variable length code is generated in the order of the divided blocks, the image is encoded within a range of a given code amount upper limit value. A code amount control method for controlling the code amount of each block for encoding,
A code amount when the block is coded based on a plurality of quantization levels for quantizing the block, and a distortion amount discarded from the block based on the quantization level are calculated. Generating a distortion / code amount table as a table;
A code amount reference value list generating step of generating a code amount reference value list including a plurality of code amount reference values having the code amount upper limit as a maximum value,
In the first block, the quantization level that is the code amount closest to the code amount reference value and the distortion amount at that time are stored in an index table and a distortion total amount table in which the code amount reference value is associated with the block. A work table initial value setting step to be set as an initial value;
For each of the code amount reference values, a quantization level for minimizing the total amount of distortion from the first block to the specific block and the total amount of distortion at that time are sequentially determined with reference to the total distortion amount table. A work table setting step of setting values of the index table and the distortion total amount table in the second and subsequent blocks,
From the index table set in this work table setting step, a reverse search step of performing a reverse search for a quantization level that minimizes the total distortion amount of the all blocks,
A code amount control method comprising: In image encoding for dividing an image to be encoded into blocks B (j) (j = 1,..., N) of a specific size and generating a variable length code for each block, a given code is used. A code amount control method for controlling the code amount of each block in order to encode the image within a range of an amount upper limit K,
The code amount R (t, j) when the block B (j) is encoded based on a plurality of quantization levels t for quantizing the block, and the block B (j) based on the quantization level t ), A distortion / code amount table generation step of calculating the amount of distortion D (t, j) to be discarded and generating a distortion / code amount table;
A code for generating a code amount reference value list including S (K) code amount reference values k (i) (i = 1,..., S (K) = K) having the code amount upper limit K as a maximum value. An amount reference value list generation step;
The quantization level t at which the code amount R (t, 1) is closest to the code amount reference value k (i) and the distortion amount D (t, 1) at that time are represented by the code amount reference value k (i). A work table initial value setting step of setting as an initial value of an index table Idx (i, j) and a total distortion amount table F (i, j) in which the table and the block B (j) are associated with each other;
The code amount is r, the mapping that returns the index i of the code amount reference value k (i) closest to r is ψ (r), and the total distortion amount f (i, j; t) is D (t, j). + F (ψ (k (i) −R (t, j)), j−1) and the quantization level t that minimizes the total distortion amount f (i, j; t), and Setting the total distortion amount f (i, j; t) in an index table Idx (i, j) and a total distortion amount table F (i, j);
S (K) is given as an index i of the code amount reference value k (i), N is given as an index j of the block B (j) as an initial value, and a quantization level t (j) assigned to the block B (j) is given. ) Is used as an index table Idx (i, j), the above-mentioned mapping ψ (k (i) −R (t (j), j)) is calculated, the calculation result is again set to the value of i, and j is sequentially subtracted. A reverse search step of performing a reverse search of the list (t (1),..., T (N)) of the quantization levels t (j) from the index table Idx (i, j);
A code amount control method comprising: The said quantization level t is a value which shows the position of the bit plane of each said block B (j), and quantizes each block B (j) based on the position of the bit plane. 3. The code amount control method according to 2. In an image encoding in which an image to be encoded is divided into blocks of a specific size and a variable length code is generated in the order of the divided blocks, the image is encoded within a range of a given code amount upper limit value. For encoding, a code amount control device that controls the code amount of each of the blocks,
A code amount when the block is coded based on a plurality of quantization levels for quantizing the block, and a distortion amount discarded from the block based on the quantization level are calculated. Distortion / code amount table generating means for generating a table,
Distortion / code amount table holding means for holding the distortion / code amount table generated by the distortion / code amount table generation means,
A code amount reference value list generating unit that generates a code amount reference value list including a plurality of code amount reference values having the code amount upper limit as a maximum value,
Code amount reference value list holding means for holding the code amount reference value list generated by the code amount reference value list generation means,
The code amount from the first block to the specific block is the quantization level of the specific block that is the code amount closest to the code amount reference value and the total distortion amount at that time, the code amount reference value and the block Work table generating means for generating an index table and a distortion total amount table in which
Work table holding means for holding the index table and the total distortion amount table generated by the work table generation means,
From the index table held in the work table holding means, a reverse search means for performing a reverse search for a quantization level that minimizes the total distortion amount of the blocks,
A code amount control device comprising: In an image encoding in which an image to be encoded is divided into blocks of a specific size and a variable length code is generated in the order of the divided blocks, the image is encoded within a range of a given code amount upper limit value. In order to control the code amount of each block when encoding, a computer,
A code amount when the block is coded based on a plurality of quantization levels for quantizing the block, and a distortion amount discarded from the block based on the quantization level are calculated. Distortion / code amount table generating means generated as a table and held in the distortion / code amount table holding means,
A code amount reference value list generation unit that generates a code amount reference value list including a plurality of code amount reference values with the code amount upper limit being the maximum value and holds the code amount reference value list holding unit,
The quantization level at which the code amount from the first block to the specific block is the code amount closest to the code amount reference value and the total distortion at that time are associated with the code amount reference value and the block. Work table generating means for generating the index table and the total distortion amount table and holding the work table holding means
From the index table held by the work table holding means, a reverse search means for performing a reverse search for a quantization level that minimizes the total distortion amount of the blocks,
A code amount control program characterized by functioning as:

Images