JP3933749B2 - Image coding apparatus and image decoding apparatus - Google Patents

Description

【００１７】
となり、画素単位では
【００１８】
【数２】

【００１９】
となる。
【００２０】
この式から、復号側で選択された画素単位の歪み量は、その画素が属するブロック内の符号化対象画素数Ｍに比例することになる。
【００２１】
伝送する情報量一定の条件で復号画像の総歪み量が最小になるよう最適化するには、すなわち、伝送情報と歪みに関して最も効率的な符号化を実現するには、画素単位の歪み量を均一にする必要がある。従って、上記のように符号化された場合、画素単位の歪み量がＭに依存してブロック毎、画像毎に異なり、最適化されないことになる。
【００２２】
そこで、本発明は上述の点に鑑みて成されたもので、上記の課題を解決した画像符号化装置および画像復号化装置を提供することを目的とする。
【００２３】
【課題を解決するための手段】
上記目的を達成するために、本発明に係る画像符号化装置の一態様は、複数の構成要素からなる画像を所定数の画素で構成される所定寸法の複数のブロックに分割し、該ブロックに含まれる単一の構成要素からなる少なくとも一つの画像領域に対応した画素信号ベクトルを、前記構成要素の境界情報であるブロック領域情報に基づいて前記少なくとも一つの画像領域毎に導出された直交変換基底ベクトルの重み付け線形和として変換出力する変換手段と、前記変換手段からの信号を量子化する量子化手段とを備え、前記直交変換基底ベクトルが、離散コサイン変換基底ベクトルの符号化対象画像領域に対応する所定成分を前記ブロック領域情報に基づいて特定した前記符号化対象画像領域毎に保持し前記所定成分以外の成分を０とするマスキング処理を施し、かつマスキングされたベクトルのノルムの自乗で除することによりスカラー倍して前記少なくとも一つの画像領域毎に新たに導出した直交変換基底ベクトルである画像符号化装置であって、前記変換手段からの重み付け変換係数を、前記ブロックに含まれる前記符号化対象画像領域毎にその画素数の平方根を前記ブロックの画素数で除した値でスカラー倍して前記量子化手段の入力とすることを特徴とする。
【００２４】
本発明に係る画像符号化装置の別の態様は、複数の構成要素からなる画像を所定数の画素で構成される所定寸法の複数のブロックに分割し、該ブロックに含まれる単一の構成要素からなる少なくとも一つの画像領域に対応した画素信号ベクトルを、前記構成要素の境界情報であるブロック領域情報に基づいて前記少なくとも一つの画像領域毎に導出された直交変換基底ベクトルの重み付け線形和として変換出力する変換手段と、前記変換手段からの信号を量子化する量子化手段とを備え、前記直交変換基底ベクトルが、離散コサイン変換基底ベクトルの符号化対象画像領域に対応する所定成分を前記ブロック領域情報に基づいて特定した前記符号化対象画像領域毎に保持し前記所定成分以外の成分を０とするマスキング処理を施し、かつマスキングされたベクトルのノルムの自乗で除することによりスカラー倍して前記少なくとも一つの画像領域毎に新たに導出した直交変換基底ベクトルである画像符号化装置であって、前記画素信号ベクトルの信号レベルを、前記ブロックに含まれる前記符号化対象画像領域毎にその画素数の平方根を前記ブロックの画素数で除した値でスカラー倍して前記変換手段の入力とすることを特徴とする。
【００２５】
本発明に係る画像符号化装置のさらに別の態様は、複数の構成要素からなる画像を所定数の画素で構成される所定寸法の複数のブロックに分割し、該ブロックに含まれる単一の構成要素からなる少なくとも一つの画像領域に対応した画素信号ベクトルを、前記構成要素の境界情報であるブロック領域情報に基づいて前記少なくとも一つの画像領域毎に導出された直交変換基底ベクトルの重み付け線形和として変換出力する変換手段と、前記変換手段からの信号を量子化する量子化手段とを備え、前記直交変換基底ベクトルが、離散コサイン変換基底ベクトルの符号化対象画像領域に対応する所定成分を前記ブロック領域情報に基づいて特定した前記符号化対象画像領域毎に保持し前記所定成分以外の成分を０とするマスキング処理を施し、かつマスキングされたベクトルのノルムの自乗で除することによりスカラー倍して前記少なくとも一つの画像領域毎に新たに導出した直交変換基底ベクトルである画像符号化装置であって、前記量子化手段は、前記ブロックに含まれる前記符号化対象画像領域毎にその画素数の平方根で前記ブロックの画素数を除した値で量子化ステップ幅をスカラー倍することを特徴とする。
【００２７】
上記目的を達成するために、本発明に係る画像復号化装置の一態様は、上記一態様の画像符号化装置から出力される、所定数の画素で構成され所定寸法で分割されたブロックに含まれる単一の構成要素からなる少なくとも一つの画像領域に対応した画素信号ベクトルを符号化した符号化データを、逆量子化して重み付け係数を求める逆量子化手段と、前記構成要素の境界情報であるブロック領域情報に基づいて前記少なくとも一つの画像領域毎に導出された直交変換基底ベクトルを重み付け線形加算して、前記少なくとも一つの画像領域に対応した画素信号ベクトルを復号する逆変換手段とを備え、前記直交変換基底ベクトルが、離散コサイン変換基底ベクトルの符号化対象画像領域に対応する所定成分を前記ブロック領域情報に基づいて特定した前記符号化対象画像領域毎に保持し前記所定成分以外の成分を０とするマスキング処理を施し、かつマスキングされたベクトルのノルムの自乗で除することによりスカラー倍して前記少なくとも一つの画像領域毎に新たに導出した直交変換基底ベクトルである画像復号化装置であって、前記逆量子化して求めた重み付け係数を、前記ブロックに含まれる前記符号化対象画像領域毎にその画素数の平方根で前記ブロックの画素数を除した値でスカラーして前記直交変換基底ベクトルを重み付け線形加算することを特徴とする。
【００２８】
本発明に係る画像復号化装置の別の態様は、上記別の態様の画像符号化装置から出力される、所定数の画素で構成され所定寸法で分割されたブロックに含まれる単一の構成要素からなる少なくとも一つの画像領域に対応した画素信号ベクトルを符号化した符号化データを、逆量子化して重み付け係数を求める逆量子化手段と、前記構成要素の境界情報であるブロック領域情報に基づいて前記少なくとも一つの画像領域毎に導出された直交変換基底ベクトルを重み付け線形加算して、前記少なくとも一つの画像領域に対応した画素信号ベクトルを復号する逆変換手段とを備え、前記直交変換基底ベクトルが、離散コサイン変換基底ベクトルの符号化対象画像領域に対応する所定成分を前記ブロック領域情報に基づいて特定した前記符号化対象画像領域毎に保持し前記所定成分以外の成分を０とするマスキング処理を施し、かつマスキングされたベクトルのノルムの自乗で除することによりスカラー倍して前記少なくとも一つの画像領域毎に新たに導出した直交変換基底ベクトルである画像復号化装置であって、前記逆変換手段出力を、前記ブロックに含まれる前記符号化対象画像領域毎にその画素数の平方根で前記ブロックの画素数を除した値でスカラー倍して前記少なくとも一つの画像領域に対応した画素信号ベクトルを復号することを特徴とする。
【００２９】
本発明に係る画像復号化装置のさらに別の態様は、上記さらに別の態様の画像符号化装置から出力される、所定数の画素で構成され所定寸法で分割されたブロックに含まれる単一の構成要素からなる少なくとも一つの画像領域に対応した画素信号ベクトルを符号化した符号化データを、逆量子化して重み付け係数を求める逆量子化手段と、前記構成要素の境界情報であるブロック領域情報に基づいて前記少なくとも一つの画像領域毎に導出された直交変換基底ベクトルを重み付け線形加算して、前記少なくとも一つの画像領域に対応した画素信号ベクトルを復号する逆変換手段とを備え、前記直交変換基底ベクトルが、離散コサイン変換基底ベクトルの符号化対象画像領域に対応する所定成分を前記ブロック領域情報に基づいて特定した前記符号化対象画像領域毎に保持し前記所定成分以外の成分を０とするマスキング処理を施し、かつマスキングされたベクトルのノルムの自乗で除することによりスカラー倍して前記少なくとも一つの画像領域毎に新たに導出した直交変換基底ベクトルである画像復号化装置であって、前記逆量子化手段は、前記ブロックに含まれる前記符号化対象画像領域毎にその画素数の平方根を前記ブロックの画素数で除した値で量子化ステップ幅をスカラー倍することを特徴とする。
【００３１】
【発明の実施の形態】
以下、図面を参照しながら本発明の実施の形態を詳細に説明する。
【００３２】
（第１の実施の形態）
図１は本発明を適用した画像符号化装置の第１の実施の形態の構成を示すブロック図である。
【００３３】
本実施の形態では、処理対象画像の画像構成要素数は２であり、分割された各単位ブロック（以下、ブロックと記す）内の画像領域は最大で２（最小は１）とする。また、各画像構成要素が画面内で占める領域を表す領域情報（境界情報）は既知であるものとする。
【００３４】
図１に示す画像符号化装置は、領域サポートＤＣＴ部１０と乗算器１２と量子化器１４と可変長符号化器１６を備えている。
【００３５】
領域サポートＤＣＴ部１０は、領域情報が入力されるブロック化回路１，処理対象の画像信号が入力されるブロック化回路２，ＤＣＴ(Discrete Cosine Transform；離散コサイン変換）基底供給回路３，サポート領域修正回路４および５，係数導出回路６ａおよび６ｂおよび６ｃ，組合せ回路９，判定回路７，並びに切替え回路８により構成されている。サポート領域修正回路は、一つの処理対象画像の画像構成要素数だけ必要である。
【００３６】
画像信号はブロック化回路２に、その領域情報はブロック化回路１に供給され、Ｎ×Ｎの一定の寸法の複数のブロックに対するブロック信号およびブロック領域情報に分割される。画素数Ｎの値は、一般的には８または１６である。
【００３７】
入力画像がたとえば人と背景の画像構成要素からなるとすると、ブロック化回路１から出力されるブロック領域情報は、各ブロックが人のみ、または背景のみを画像構成要素とすること、或いは人の一部と背景の一部を画像構成要素とすることを表す。換言すれば、これらのブロック領域情報は、ブロック内に画像構成要素の境界がなく画像が矩形で構成されること、或いはブロック内に画像構成要素の境界があり各画像が非矩形で構成されることを表す。
【００３８】
このようにブロック化回路１によってブロック化されたブロック領域情報は、サポート領域修正回路４および５，係数導出回路６ａ，並びに判定回路７に供給される。サポート領域修正回路４および５では、ＤＣＴ基底供給回路３から供給されたＤＣＴ基底のサポート領域をブロック毎に画像領域形状に応じて修正する。サポート領域修正回路４では２つの画像領域のうち１つの画像領域（たとえば人とする）の領域形状に、サポート領域修正回路５ではもう一方の画像領域（背景とする）の領域形状に対応して修正が行われる。サポート領域修正回路４および５によるサポート領域修正方法については後述する。
【００３９】
また、ブロック化回路２によってブロック化された画像信号は、係数導出回路６ａおよび６ｂおよび６ｃに供給される。人のみで構成されるブロックおよび背景のみで構成されるブロックの画像信号に対しては、係数導出回路６ａによりＤＣＴ基底供給回路３から供給された２次元ＤＣＴ基底を修正せずに用いて、ＤＣＴ係数の導出およびその符号化を行う。係数導出回路６ａからの符号化データは、切替え回路８に供給される。
【００４０】
２つの画像構成要素のそれぞれの一部が混在するブロックの画像信号に対しては、係数導出回路６ｂおよび６ｃにより各々サポート領域修正回路４および５から供給される修正されたＤＣＴ基底を用いてＤＣＴ係数の導出およびその符号化を行う。このＤＣＴ係数の導出およびその符号化については後述する。
【００４１】
係数導出回路６ｂおよび６ｃからの符号化データは、それぞれ組み合わせ回路９に供給される。そして、１つのブロックを構成する２つの画像領域を記述する符号化データが多重された符号化データとされて、組み合わせ回路９から切替え回路８に供給される。組み合わせ回路９による符号化データの多重は、切替え回路８によりブロック単位の切替えを行なうために１つのブロック内の２つの画像領域の符号化データをまとめるための処理であり、係数導出回路６ｂおよび６ｃからの２つの符号化データ列を単に直列に並べるものである。
【００４２】
一方、判定回路７では、ブロック化回路１から供給された各ブロックのブロック領域情報に基づいて、各ブロック毎に１つの画像領域（人または背景）のみが存在するか２つの画像領域（人および背景）が混在するかを判定する。この判定回路７からの判定信号は、切替え回路８に供給される。
【００４３】
切替え回路８は判定回路７からの判定信号に基づき、１つの画像領域のみが存在するブロックに対しては係数導出回路６ａからの符号化データを、２つの画像領域が混在するブロックに対しては組み合わせ回路９からの符号化データを選択的に切替えて圧縮された直列データとして出力する。すなわち、２つの画像領域が混在するブロックに対しては組み合わせ回路９からの直列に多重された符号化データを出力し、その他の１つの画像領域から構成されるブロックに対しては係数導出回路６ａからの重み付け線形和で表される符号化データを出力する。
【００４４】
ここで、サポート領域修正回路４および５によるサポート領域修正の方法と、係数導出回路６ｂおよび６ｃにおける係数導出方法について簡単に説明する。
【００４５】
サポート領域修正では、ＤＣＴ基底供給回路３から供給される予め定められた大きさＮ×Ｎの
【００４６】
【外１】

【００４７】
が供給される。
【００４８】
上記ＤＣＴ直交変換基底ベクトルのＮ×Ｎ個の各々の成分のうち、符号化対象（サポート）画像領域に対応するＭ個の成分を保持し、その他の成分が０となるように
【００４９】
【外２】

【００５０】
を設定し、符号化対象画像領域で修正する。そして、以下に示す式（１）の演算を行って
【００５１】
【外３】

【００５２】
を求め、マスキングされたベクトルをノルムの自乗で除算することで符号化対象画像構成要素の形状に修正された新たな変換基底ベクトルを算出し、ノルム補正を行う。すなわち、
【００５３】
【数３】

【００５４】
により
【００５５】
【外４】

【００５６】
が求められる。ノルム補正後の画像構成要素の形状に修正された新たな変換基底ベクトルは、マスキングされたベクトルの符号化対象画素に対応する各成分をスカラー倍したものとなっている。
【００５７】
次に、係数導出回路７および８における係数導出法について説明する。
【００５８】
ここで、
【００５９】
【外５】

【００６０】
を示すブロック化回路２からの画像信号ベクトル（要素数Ｎ×Ｎ）とする。
【００６１】
まず、第１の段階では、上記画像信号ベクトルとノルム補正後の修正された
【００６２】
【外６】

【００６３】
との内積を計算してスカラー関数
【００６４】
【数４】

【００６５】
に変換して修正された
【００６６】
【外７】

【００６７】
を求め、これらの係数の中から次式により絶対値が最大の係数ｃ_k を選択する。
【００６８】
【数５】

【００６９】
次に、第２の段階では、絶対値最大の係数ｃ_k に対応する
【００７０】
【外８】

【００７１】
次式のとおり絶対値が最大の係数ｃ_k と絶対値が最大の係数ｃ_k に対応する上記ＤＣＴ直交変換基底ベクトルとの積の成分を上記画像信号ベクトルから減じて除去することで残差信号ベクトルを求め、これを画像信号ベクトルと置き換える。
【００７２】
【数６】

【００７３】
次に、第３の段階では、第２の段階において求めた上記の残差信号ベクトルで置き換えられた画像信号ベクトルと前記したノルム補正後の修正された新たな変換基底ベクトルとの内積を第１の段階と同様に再び計算してスカラー関数に変換して、この関数の係数のうちから第１の段階と同様に新たな絶対値最大の係数を選択する。
【００７４】
次に、第４の段階では、第３の段階において選択された絶対値が最大の係数ｃ_k とこの係数に対応するＤＣＴ直交変換基底ベクトルとの積の成分を画像信号ベクトルから減じて除去することで第２の段階と同様に再び新たな残差信号ベクトルを求め、画像信号ベクトルと置き換える。
【００７５】
次に、第５の段階では、第４の段階において求められた残差信号ベクトル、すなわち画像信号ベクトルに基づいて、残差信号ベクトルによる残差信号電力が予め十分小さな値に定められた閾値以下であるかを評価し、閾値以下であればこのときの修正された新たな変換基底ベクトルの係数を採用し、重み付け線形和として符号化して変換出力する。ここで残差信号電力の評価は符号化対象画像構成要素分だけとする。すなわち、第２の段階での処理はＮ×Ｎの要素に対して施されるが、目的は符号化対象のＭ個の画素のレベルを表現することにあるので、誤差の評価は残差信号ベクトルの要素のうち、このＭ個に対応する要素の自乗和で行なう。
【００７６】
一方、残差信号電力が十分小さな閾値以下になっていないと判定されたときは再び第３の段階に戻って処理を続行し、第５の段階において残差信号電力が予め十分小さな値に定められた閾値以下であると評価されるまで、第３の段階と第４の段階と第５の段階の処理を繰り返し実行する。
【００７７】
なお、各変換基底ベクトルは直交していないので、一度除かれた成分が他の成分の除去に伴って再び現れることもある。この場合、同一変換基底の係数を累積し、その結果を最終的に採用する。
【００７８】
このように、ＲＳ−ＤＣＴ部１０によれば、比較的小さなハードウェア規模で、高能率かつ画像構成要素単位で画像信号のハンドリングを行うことができる。
【００７９】
ＲＳ−ＤＣＴ部１０により導出された（Ｎ×Ｎ）個の変換係数は、乗算器１２に供給され、
【００８０】
【外９】

【００８１】
量子化器１４に供給され、量子化インデックス信号が出力される。出力されたインデックス信号は可変長符号化器１６に供給されて２値の符号化データ列に変換され、通信メディアや記録メディア等の伝送メディアを介して伝送される。
【００８２】
このように符号化すると、１個の符号化対象画素の量子化雑音電力は量子化前のスカラー倍の係数にかかわらずＰｅで表され、量子化雑音電力によるブロック全体の符号化歪みはＭ×Ｐｅで表される。
【００８３】
伝送メディアからの符号化データ列は、図２に示す画像復号化装置により復号される。
【００８４】
この復号処理において、符号化データ列が可変長復号化器２０に供給され、復号量子化インデックスが出力される。出力された復号量子化インデックスは逆量子化器２２に供給され、逆量子化変換係数が出力される。この逆量子化変換係数は
【００８５】
【外１０】

【００８６】
つまり、符号化画素数Ｍの平方根に逆比例した値でスカラー倍される。
【００８７】
ここで、各画像構成要素（オブジェクト）が画面内で占める領域を表す領域情報（境界情報）は既知であり、各ブロック内の符号化対象画素数ＭはＤＣＴ係数を導出する際に既に求められているものを用いる。
【００８８】
スカラー倍された逆量子化変換係数はＮ×Ｎの２次元逆ＤＣＴ部２６に供給され、（Ｎ×Ｎ）個の画素信号レベルが復元される。このように、
【００８９】
【外１１】

【００９０】
復元された画素信号のブロック全体での量子化雑音電力による歪みは、
【００９１】
【数７】

【００９２】
となる。
【００９３】
復元された（Ｎ×Ｎ）個の画素信号は選択器２８に供給される。選択器２８においては、別途伝送された当該ブロックにおける画像領域を示すマスクパターン２９を用いて、（Ｎ×Ｎ）個のブロック内全画素からブロックの画像領域内に対応するＭ個の画素を選択し、出力する。この結果、ブロック全体での最終的な歪み量は
【００９４】
【数８】

【００９５】
となる。
【００９６】
従って、画素単位での量子化雑音による符号化歪みの量は符号化対象画素数Ｍに逆比例するので、画素単位での最終的な歪み量は、ＲＳ−ＤＣＴにおける符号化対象画素数Ｍとの比例性を相殺し、Ｍに依存することなく一律にＰｅとなり、画素単位の歪み量を均一に最適化することができる。
【００９７】
このように、画像符号化装置において量子化器の前に乗算器１２を設けて
【００９８】
【外１２】

【００９９】
復号した画素単位の歪み量が均一となるように情報量割り当てを最適化して、伝送する情報量を一定とした場合の復号画像の総歪み量を最小とすることができ、より高能率な符号化データの伝送を実現できる効果がある。
【０１００】
（他の実施の形態）
上記実施の形態における各乗算器１２，２４による処理はスカラー倍であり、これは線形変換に対して不変である。従って、乗算器によるこのスカラー倍を、画素信号レベルに対して行っても、上記実施の形態において変換係数に対して行ったのと等価になる。
【０１０１】
そこで、符号化装置においてブロック化回路２の入力または出力に乗算器を設け、ここでの画素信号レベルに対して図１の乗算器１２と同様のスカラー倍を行い、復号化装置では２次元逆ＤＣＴ部２６の出力に乗算器を設け、ここでの画素信号レベルに対して図２の乗算器２４と同様のスカラー倍を行っても第１の実施の形態と同様の効果を得ることができる。
【０１０２】
また、乗算器を設けることなく、量子化雑音電力が量子化ステップ幅の二乗に比例して増加することを利用して画素単位の歪み量を均一に最適化し、第１の実施の形態と同様の効果を得るように構成することもできる。
【０１０３】
すなわち、符号化装置において量子化器１４の量子化ステップ幅をＲＳ−ＤＣＴ部１０における符号化対象画素数Ｍの平方根に逆比例した値でスカラー倍して修正し、さらに、復号化装置では逆量子化器２２の量子化ステップ幅を符号化対象画素数Ｍの平方根に比例した値でスカラー倍して修正することによっても等価な効果が得られる。
【０１０４】
ただし、量子化処理は乗算とは逆の除算なので、符号化装置における量子化ステップ幅は、第１の実施の形態において
【０１０５】
【外１３】

【０１０６】
を用いてスカラー倍する必要がある。
【０１０７】
【発明の効果】
以上説明してきたように、本発明によれば、複数の構成要素からなる画像の境界情報であるブロック領域情報に基づく符号化処理を施した符号化装置から出力される符号化データを復号化装置により復号化するときに、復号画像の総歪み量を最小とするように復号化処理を施すことができる効果が得られる。
【図面の簡単な説明】
【図１】本発明を適用した画像符号化装置の第１の実施の形態の構成を示すブロック図である。
【図２】本発明を適用した画像復号化装置の第１の実施の形態の構成を示すブロック図である。
【符号の説明】
１，２ ブロック化回路
３ ＤＣＴ基底供給回路
４，５ サポート領域修正回路
６ａ，６ｂ，６ｃ 係数導出回路
７ 判定回路
８ 切替え回路
９ 組み合わせ回路
１０ ＲＳ−ＤＣＴ部
１２，２４ 乗算器
１４ 量子化器
１６ 可変長符号化器
２０ 可変長復号化器
２２ 逆量子化器
２６ ２次元逆ＤＣＴ部
２８ 選択器
２９ マスクパターン[0001]
BACKGROUND OF THE INVENTION
The present invention relates to an image encoding device and a decoding device, and more particularly to an image encoding device that performs high-efficiency encoding by processing an image signal such as a television signal in units of image components and a medium. The present invention relates to an image decoding apparatus that decodes encoded data from an encoding apparatus.
[0002]
[Prior art]
In conventional high-efficiency coding of image signals, the signal power is concentrated on a specific transform coefficient by orthogonal transform using the horizontal or vertical correlation of the image signal to reduce the required number of bits. Many. In this high-efficiency encoding using orthogonal transform, one image is divided into a plurality of unit blocks each having a fixed number of pixels, and conversion is performed for each block, and the pixels in this block are used as a basic basis. This is expressed as a weighted linear sum of vectors, the weights are quantized, and the corresponding numbers are transmitted via the media.
[0003]
In conventional image coding represented by MPEG standardization, a single image is treated only as a continuous data string regardless of the image content, and the waveform (pattern) of the image is more faithful with a smaller number of bits. The goal is to reproduce. In such a conventional coding, since processing such as DCT (Discrete Cosine Transform) that is basically based on steadyness is usually used, the non-stationary nature inherent in the image signal such as the foreground and the background is used. This leads to a decrease in encoding efficiency and decoded image quality. Conventionally, image signal processing and handling are collectively handled for each image, or in the case of moving images, for each continuous frame. Therefore, it has not been considered to perform signal processing or handling in smaller units such as image component units.
[0004]
Therefore, in recent years, research on coding in which an image signal is processed and coded after being divided into a plurality of regions corresponding to each image constituent element has become active. In this encoding, since the above-mentioned non-stationarity can be avoided and an optimum processing method can be selected and applied for each region, higher-performance encoding can be performed. In addition, processing in units of image components is possible, and video description with better handling can be realized.
[0005]
In the conventional waveform coding of an image signal, an image is divided into rectangular blocks of a certain size and processed as represented by DCT coding.
[0006]
However, when processing is performed for each area by dividing into a plurality of areas, the area shape generally does not become a rectangle, so the conventional method of dividing an image into rectangular blocks of a certain size cannot be used as it is. Therefore, several extended methods for processing a pixel group having an arbitrary shape have been proposed based on the conventional block-based encoding method.
[0007]
As one of such methods, the DCT transform base is masked by the image area of the encoding target image and is multiplied by a scalar to derive a new transform basis vector for each image area, and conversion is performed using this. A region support DCT (RS-DCT) for obtaining a coefficient has been proposed by the present applicants (Yoshiaki Shikaga, “Examination of Arbitrary Shape Coding Using Region Support DCT”) IEICE Tech. 109 (Dec. 1995), Japanese Patent Application No. 7-325971). The transform coefficient derived thereby can be decoded by a well-known two-dimensional inverse DCT.
[0008]
As a technique similar to this RS-DCT, in order to use a rectangular two-dimensional DCT, a pixel that does not belong to the encoding target area within the rectangle is extrapolated by, for example, averaging the levels of the pixels in the area. RF-DCT (Region Filling DCT) has also been proposed (Keiichi Kimura: “Examination of Image Coding by Intrablock Region Division” Proc. PCSJ, '95, 3-7, pp. 39-40 (Oct. 1995)). Norio Ito, Hiroyuki Katada, Hiroshi Kusao: “Examination of DCT Realization Method for Arbitrary Shape Regions”, Proc. PCSJ, '95, 5-2, pp. 77-78 (Oct. 1995)).
[0009]
In RS-DCT and RF-DCT, for example, when two image areas coexist in a unit block, block DCT is performed for both image areas. For this reason, the transform coefficient is derived for each image area by the size of the block (ie, the number of pixels of the block including the two image areas). Therefore, when the transform coefficients for both image regions are combined, two blocks of coefficients are derived for one block, and it is necessary to transmit the transform coefficients for these two blocks. Thus, in RS-DCT and RF-DCT, when two image regions are mixed in a block, it is necessary to transmit a conversion coefficient that is apparently twice the number of pixels. However, in normal high-efficiency coding using DCT, only the transform coefficients corresponding to the low-frequency components that are visually important and occupy most of the signal power are transmitted, so that the number of transform coefficients to be transmitted is apparently increased. This does not lead to an effective increase in the amount of transmitted information.
[0010]
In these methods, the transform coefficient is derived from the encoding target pixel included in the block, but on the decoding side, the block-shaped image is apparently restored by the two-dimensional inverse DCT, and the encoding target region that is separately transmitted is Only the pixels in the corresponding region are selected by the mask pattern shown and used for the decoded image.
[0011]
Although the above method is simple, continuity in the low-frequency region and its vicinity is guaranteed to some extent, but it becomes discontinuous in the middle and high-frequency regions, and is optimal as a pixel description method for the inserted pixel level and thus in the region. There is no guarantee of sex.
[0012]
[Problems to be solved by the invention]
In both RS-DCT and RF-DCT, the number of transform bases matches the number of pixels in the block (hereinafter referred to as N × N) regardless of the number of pixels to be encoded (hereinafter referred to as M). Therefore, apparently (N × N) conversion coefficients are obtained. However, since the conversion is performed on the values of M (M ≦ N × N) encoding target pixels, the actually significant (non-zero) conversion coefficients obtained are at most M. Only.
[0013]
In high-efficiency coding using DCT, the transform coefficient is quantized to reduce the amount of information, and at the same time, coding distortion due to quantization error during quantization occurs. However, the quantization noise is 0 for a pixel having a transform coefficient value of 0. Accordingly, in the entire block including M encoding target pixels, the quantization noise is M. That is, assuming that the quantization noise power for one pixel to be encoded is Pe, M significant transform coefficients are obtained for M pixels, and as a result, M × Pe encoding distortion is applied to the block. Bring.
[0014]
After a block-like decoded image of (N × N) pixels is once obtained from the encoded data obtained by such DCT by two-dimensional inverse DCT, only M of these pixels are mask patterns. Is left selected. Along with this processing, distortion due to quantization noise distributed to other than the selected pixel is also removed.
[0015]
That is, the distortion caused by the quantization noise left in the block is
[0016]
[Expression 1]

[0017]
And in pixel units
[0018]
[Expression 2]

[0019]
It becomes.
[0020]
From this equation, the amount of distortion in pixel units selected on the decoding side is proportional to the number M of pixels to be encoded in the block to which the pixel belongs.
[0021]
In order to optimize the total distortion amount of the decoded image under the condition that the amount of information to be transmitted is constant, that is, in order to achieve the most efficient encoding with respect to transmission information and distortion, the distortion amount per pixel is set. It needs to be uniform. Therefore, when encoding is performed as described above, the amount of distortion in pixel units depends on M and differs from block to block and from image to image, and is not optimized.
[0022]
Therefore, the present invention has been made in view of the above points, and an object thereof is to provide an image encoding device and an image decoding device that have solved the above problems.
[0023]
[Means for Solving the Problems]
In order to achieve the above object, an aspect of an image encoding device according to the present invention divides an image composed of a plurality of components into a plurality of blocks each having a predetermined size including a predetermined number of pixels. An orthogonal transform base derived for each at least one image area based on block area information, which is boundary information of the constituent elements, of a pixel signal vector corresponding to at least one image area composed of a single constituent element A conversion unit that converts and outputs a weighted linear sum of vectors; and a quantization unit that quantizes a signal from the conversion unit, wherein the orthogonal transform base vector corresponds to an encoding target image region of the discrete cosine transform base vector A predetermined component to be stored for each of the encoding target image regions specified based on the block region information, and a component other than the predetermined component to be zero. Applying grayed treatment, and the square of the norm of the masked vector By dividing by An image encoding device which is an orthogonal transform base vector newly derived for each of the at least one image region after being multiplied by a scalar, wherein the weighted transform coefficient from the transform unit is used as the encoding target image included in the block. For each region, it is scalar-multiplied by a value obtained by dividing the square root of the number of pixels by the number of pixels of the block, and used as the input of the quantization means.
[0024]
Another aspect of the image coding apparatus according to the present invention divides an image composed of a plurality of components into a plurality of blocks having a predetermined size composed of a predetermined number of pixels, and a single component included in the block A pixel signal vector corresponding to at least one image area consisting of is converted as a weighted linear sum of orthogonal transform base vectors derived for each of the at least one image area based on block area information which is boundary information of the constituent elements A transforming means for outputting; and a quantizing means for quantizing a signal from the transforming means, wherein the orthogonal transform basis vector represents a predetermined component corresponding to an encoding target image region of a discrete cosine transform basis vector in the block region. A masking process is performed for each of the encoding target image areas specified based on the information, and a component other than the predetermined component is set to 0. The square of the norm of the ring has been vector By dividing by An image encoding device that is an orthogonal transform base vector newly derived for each of the at least one image area after being multiplied by a scalar, wherein the signal level of the pixel signal vector is set to the image area to be encoded included in the block Each time, the square root of the number of pixels divided by the number of pixels in the block is multiplied by a scalar to be input to the conversion means.
[0025]
Still another aspect of the image encoding device according to the present invention is to divide an image composed of a plurality of components into a plurality of blocks having a predetermined size composed of a predetermined number of pixels, and to provide a single configuration included in the block A pixel signal vector corresponding to at least one image region made up of elements is used as a weighted linear sum of orthogonal transformation base vectors derived for each of the at least one image region based on block region information that is boundary information of the constituent elements. Transform means for transforming and quantizing means for quantizing a signal from the transform means, wherein the orthogonal transform base vector is a block that outputs a predetermined component corresponding to an encoding target image region of a discrete cosine transform base vector. A masking process is performed for each of the encoding target image areas specified based on the area information, and a component other than the predetermined component is set to 0. The square of the norm of the masked vector By dividing by An image encoding device, which is an orthogonal transform basis vector newly derived for each of the at least one image region after being multiplied by a scalar, wherein the quantizing means includes an encoding unit for each of the encoding target image regions included in the block. The quantization step width is scalar multiplied by a value obtained by dividing the number of pixels of the block by the square root of the number of pixels.
[0027]
In order to achieve the above object, one aspect of an image decoding apparatus according to the present invention is included in a block composed of a predetermined number of pixels and divided by a predetermined size, which is output from the image encoding apparatus of the above aspect. Dequantization means for inversely quantizing encoded data obtained by encoding a pixel signal vector corresponding to at least one image region composed of a single constituent element to obtain a weighting coefficient, and boundary information of the constituent element Inverse conversion means for decoding the pixel signal vector corresponding to the at least one image area by performing weighted linear addition of the orthogonal transform base vectors derived for each of the at least one image area based on the block area information, Based on the block region information, the orthogonal transform basis vector has a predetermined component corresponding to the encoding target image region of the discrete cosine transform basis vector. And said encoding target image is stored for each area component other than the predetermined component masked process to 0, and the square of the norm of the masked vector By dividing by An image decoding device, which is an orthogonal transform basis vector newly derived for each of the at least one image region after being multiplied by a scalar, wherein the weighting coefficient obtained by the inverse quantization is the encoding target included in the block For each image area, the orthogonal transformation base vector is weighted linearly added and scalarized by a value obtained by dividing the number of pixels of the block by the square root of the number of pixels.
[0028]
Another aspect of the image decoding apparatus according to the present invention is a single component included in a block composed of a predetermined number of pixels and divided by a predetermined size, which is output from the image encoding apparatus according to another aspect. Inverse quantization means for inversely quantizing encoded data obtained by encoding a pixel signal vector corresponding to at least one image area and obtaining a weighting coefficient, and block area information which is boundary information of the constituent elements Inverse transformation means for decoding the pixel signal vector corresponding to the at least one image area by performing weighted linear addition of the orthogonal transformation basis vectors derived for each of the at least one image area, The encoding target image in which a predetermined component corresponding to the encoding target image region of the discrete cosine transform basis vector is specified based on the block region information. It was held for each domain components other than the predetermined component masked process to 0, and the square of the norm of the masked vector By dividing by An image decoding apparatus that is an orthogonal transform basis vector newly derived for each of the at least one image area after being multiplied by a scalar, wherein the inverse transform means output is output for each of the encoding target image areas included in the block. The pixel signal vector corresponding to the at least one image area is decoded by multiplying by a value obtained by dividing the number of pixels of the block by the square root of the number of pixels.
[0029]
Still another aspect of the image decoding apparatus according to the present invention is a single block included in a block composed of a predetermined number of pixels and divided by a predetermined size, which is output from the image encoding apparatus of the still another aspect. Inverse quantization means for inversely quantizing encoded data obtained by encoding a pixel signal vector corresponding to at least one image area composed of constituent elements to obtain weighting coefficients, and block area information which is boundary information of the constituent elements Inverse transform means for decoding the pixel signal vector corresponding to the at least one image region by weighted linearly adding the orthogonal transform basis vectors derived for each of the at least one image region based on the orthogonal transform base, The vector specifies a predetermined component corresponding to an encoding target image region of a discrete cosine transform base vector based on the block region information No. of the target image held in each region subjected to masking to 0 the component other than the predetermined component and the square of the norm of the masked vector By dividing by An image decoding device that is an orthogonal transform basis vector newly derived for each of the at least one image region after being multiplied by a scalar, wherein the inverse quantization means is provided for each of the encoding target image regions included in the block. The quantization step width is scalar multiplied by a value obtained by dividing the square root of the number of pixels by the number of pixels of the block.
[0031]
DETAILED DESCRIPTION OF THE INVENTION
Hereinafter, embodiments of the present invention will be described in detail with reference to the drawings.
[0032]
(First embodiment)
FIG. 1 is a block diagram showing a configuration of a first embodiment of an image encoding apparatus to which the present invention is applied.
[0033]
In the present embodiment, the number of image components of the processing target image is 2, and the maximum image area in each divided unit block (hereinafter referred to as a block) is 2 (minimum is 1). Further, it is assumed that area information (boundary information) representing an area occupied by each image component in the screen is known.
[0034]
The image encoding apparatus illustrated in FIG. 1 includes a region support DCT unit 10, a multiplier 12, a quantizer 14, and a variable length encoder 16.
[0035]
The region support DCT unit 10 includes a blocking circuit 1 to which region information is input, a blocking circuit 2 to which an image signal to be processed is input, a DCT (Discrete Cosine Transform) base supply circuit 3, and a support region correction. The circuits 4 and 5, the coefficient deriving circuits 6a and 6b and 6c, the combinational circuit 9, the determination circuit 7, and the switching circuit 8 are included. The support area correction circuit is required for the number of image components of one processing target image.
[0036]
The image signal is supplied to the blocking circuit 2, and the area information is supplied to the blocking circuit 1, and is divided into block signals and block area information for a plurality of blocks having a fixed size of N × N. The value of the number of pixels N is generally 8 or 16.
[0037]
If the input image is made up of, for example, human and background image components, the block area information output from the blocking circuit 1 is that each block has only a person or only the background as an image component, or part of a person. And that part of the background is an image component. In other words, the block area information is such that there is no image component boundary in the block and the image is rectangular, or each block has image component boundaries and each image is non-rectangular. Represents that.
[0038]
The block area information blocked by the blocking circuit 1 in this way is supplied to the support area correction circuits 4 and 5, the coefficient derivation circuit 6a, and the determination circuit 7. The support area correction circuits 4 and 5 correct the DCT base support area supplied from the DCT base supply circuit 3 according to the image area shape for each block. The support area correction circuit 4 corresponds to the area shape of one of the two image areas (eg, a person), and the support area correction circuit 5 corresponds to the area shape of the other image area (ie, the background). Corrections are made. A support area correction method by the support area correction circuits 4 and 5 will be described later.
[0039]
The image signal blocked by the blocking circuit 2 is supplied to the coefficient deriving circuits 6a, 6b and 6c. For image signals of a block composed only of people and a block composed only of background, the two-dimensional DCT base supplied from the DCT base supply circuit 3 by the coefficient derivation circuit 6a is used without modification, and the DCT is used. Coefficients are derived and encoded. The encoded data from the coefficient derivation circuit 6a is supplied to the switching circuit 8.
[0040]
For an image signal of a block in which a part of each of the two image components is mixed, DCT is performed using the modified DCT bases supplied from the support region correction circuits 4 and 5 by the coefficient derivation circuits 6b and 6c, respectively. Coefficients are derived and encoded. The derivation of the DCT coefficient and its encoding will be described later.
[0041]
The encoded data from the coefficient derivation circuits 6b and 6c are supplied to the combinational circuit 9, respectively. Then, the encoded data describing the two image areas constituting one block is encoded data and supplied from the combinational circuit 9 to the switching circuit 8. The multiplexing of the encoded data by the combinational circuit 9 is a process for collecting the encoded data of two image areas in one block in order to switch the block unit by the switching circuit 8, and the coefficient deriving circuits 6b and 6c. Are simply arranged in series.
[0042]
On the other hand, in the determination circuit 7, based on the block area information of each block supplied from the block forming circuit 1, there is only one image area (person or background) for each block, or two image areas (person and background). It is determined whether background is mixed. The determination signal from the determination circuit 7 is supplied to the switching circuit 8.
[0043]
Based on the determination signal from the determination circuit 7, the switching circuit 8 applies the encoded data from the coefficient derivation circuit 6a to a block in which only one image area exists, and to the block in which two image areas are mixed. The encoded data from the combinational circuit 9 is selectively switched and output as compressed serial data. That is, the encoded data multiplexed in series from the combinational circuit 9 is output for a block in which two image areas are mixed, and the coefficient deriving circuit 6a for a block composed of one other image area. The encoded data represented by the weighted linear sum from is output.
[0044]
Here, a method of correcting the support area by the support area correction circuits 4 and 5 and a coefficient derivation method in the coefficient derivation circuits 6b and 6c will be briefly described.
[0045]
In the support area correction, a predetermined size N × N supplied from the DCT base supply circuit 3 is used.
[0046]
[Outside 1]

[0047]
Is supplied.
[0048]
Of the N × N components of the DCT orthogonal transform basis vector, M components corresponding to the encoding target (support) image region are held, and the other components are zero.
[0049]
[Outside 2]

[0050]
Is corrected in the encoding target image area. Then, the following equation (1) is calculated
[0051]
[Outside 3]

[0052]
Then, the masked vector is divided by the square of the norm to calculate a new transform base vector corrected to the shape of the encoding target image constituent element, and the norm correction is performed. That is,
[0053]
[Equation 3]

[0054]
By
[0055]
[Outside 4]

[0056]
Is required. The new transform base vector corrected to the shape of the image component after the norm correction is obtained by multiplying each component corresponding to the pixel to be encoded of the masked vector by scalar multiplication.
[0057]
Next, a coefficient derivation method in the coefficient derivation circuits 7 and 8 will be described.
[0058]
here,
[0059]
[Outside 5]

[0060]
Is an image signal vector (number of elements N × N) from the blocking circuit 2.
[0061]
First, in the first stage, the image signal vector and the norm corrected image are corrected.
[0062]
[Outside 6]

[0063]
A scalar function
[0064]
[Expression 4]

[0065]
Converted to
[0066]
[Outside 7]

[0067]
From these coefficients, the coefficient c having the maximum absolute value is _k Select.
[0068]
[Equation 5]

[0069]
Next, in the second stage, the coefficient c having the maximum absolute value _k Corresponding to
[0070]
[Outside 8]

[0071]
Coefficient c with the largest absolute value as _k And the coefficient c with the maximum absolute value _k A residual signal vector is obtained by subtracting and removing the product component of the DCT orthogonal transform base vector corresponding to the image signal vector from the image signal vector, and replacing this with the image signal vector.
[0072]
[Formula 6]

[0073]
Next, in the third stage, the inner product of the image signal vector replaced with the residual signal vector obtained in the second stage and the new transformed basis vector corrected after the norm correction is used as the first product. In the same manner as in step (1), the calculation is performed again and converted into a scalar function, and a new coefficient having the maximum absolute value is selected from the coefficients of the function in the same manner as in the first step.
[0074]
Next, in the fourth stage, the absolute value selected in the third stage is the maximum coefficient c. _k Then, the product component of the coefficient and the DCT orthogonal transform base vector corresponding to this coefficient is subtracted from the image signal vector and removed, so that a new residual signal vector is obtained again in the same manner as in the second stage and replaced with the image signal vector.
[0075]
Next, in the fifth stage, based on the residual signal vector obtained in the fourth stage, that is, the image signal vector, the residual signal power based on the residual signal vector is equal to or less than a threshold value set to a sufficiently small value in advance. If it is equal to or less than the threshold value, the coefficient of the new conversion basis vector corrected at this time is adopted, encoded as a weighted linear sum, and converted and output. Here, the residual signal power is evaluated only for the constituent elements of the encoding target image. That is, the processing in the second stage is performed on N × N elements, but the purpose is to express the level of M pixels to be encoded. Of the vector elements, the sum of squares of the elements corresponding to the M elements is used.
[0076]
On the other hand, when it is determined that the residual signal power is not below a sufficiently small threshold value, the process returns to the third stage to continue the process, and the residual signal power is set to a sufficiently small value in the fifth stage in advance. The processes of the third stage, the fourth stage, and the fifth stage are repeatedly executed until it is evaluated that it is equal to or less than the threshold value.
[0077]
In addition, since each conversion base vector is not orthogonal, the component once removed may reappear with the removal of another component. In this case, the coefficients of the same conversion base are accumulated, and the result is finally adopted.
[0078]
As described above, according to the RS-DCT unit 10, it is possible to handle an image signal with high efficiency and in units of image components with a relatively small hardware scale.
[0079]
The (N × N) transform coefficients derived by the RS-DCT unit 10 are supplied to the multiplier 12,
[0080]
[Outside 9]

[0081]
The quantized signal is supplied to the quantizer 14 and a quantized index signal is output. The output index signal is supplied to the variable length encoder 16 and converted into a binary encoded data string, and transmitted via a transmission medium such as a communication medium or a recording medium.
[0082]
When encoding is performed in this way, the quantization noise power of one pixel to be encoded is represented by Pe regardless of the scalar multiplication coefficient before quantization, and the coding distortion of the entire block due to the quantization noise power is M ×. Represented by Pe.
[0083]
The encoded data string from the transmission medium is decoded by the image decoding apparatus shown in FIG.
[0084]
In this decoding process, the encoded data string is supplied to the variable length decoder 20 and a decoded quantization index is output. The output decoded quantization index is supplied to the inverse quantizer 22 and the inverse quantization transform coefficient is output. This inverse quantization transform coefficient is
[0085]
[Outside 10]

[0086]
That is, it is scalar multiplied by a value inversely proportional to the square root of the number M of encoded pixels.
[0087]
Here, the area information (boundary information) representing the area occupied by each image component (object) in the screen is known, and the number of encoding target pixels M in each block is already obtained when the DCT coefficient is derived. Use what you have.
[0088]
The scalar multiplied inverse quantization transform coefficient is supplied to the N × N two-dimensional inverse DCT unit 26, and (N × N) pixel signal levels are restored. in this way,
[0089]
[Outside 11]

[0090]
Distortion due to quantization noise power in the entire block of the restored pixel signal is
[0091]
[Expression 7]

[0092]
It becomes.
[0093]
The restored (N × N) pixel signals are supplied to the selector 28. In the selector 28, M pixels corresponding to the image area of the block are selected from all the pixels in the (N × N) blocks using the mask pattern 29 indicating the image area of the block transmitted separately. And output. As a result, the final amount of distortion in the entire block is
[0094]
[Equation 8]

[0095]
It becomes.
[0096]
Accordingly, since the amount of encoding distortion due to quantization noise in pixel units is inversely proportional to the number of encoding target pixels M, the final amount of distortion in pixel units is the number of encoding target pixels M in RS-DCT. The proportionality is canceled out and becomes Pe uniformly without depending on M, and the distortion amount in pixel units can be optimized uniformly.
[0097]
As described above, in the image encoding apparatus, the multiplier 12 is provided before the quantizer.
[0098]
[Outside 12]

[0099]
By optimizing the information amount allocation so that the decoded pixel unit distortion amount is uniform, the total distortion amount of the decoded image can be minimized when the amount of information to be transmitted is constant, and a more efficient code There is an effect that can realize the transmission of digitized data.
[0100]
(Other embodiments)
The processing by each of the multipliers 12 and 24 in the above embodiment is a scalar multiplication, which is invariant to linear transformation. Therefore, even if this scalar multiplication by the multiplier is performed on the pixel signal level, it is equivalent to that performed on the conversion coefficient in the above embodiment.
[0101]
Therefore, a multiplier is provided at the input or output of the blocking circuit 2 in the encoding device, and the pixel signal level here is multiplied by the same scalar multiplication as the multiplier 12 in FIG. 1, and the two-dimensional inverse in the decoding device. Even if a multiplier is provided at the output of the DCT unit 26 and the pixel signal level here is multiplied by the same scalar multiplication as that of the multiplier 24 of FIG. 2, the same effect as in the first embodiment can be obtained. .
[0102]
Also, without providing a multiplier, the distortion amount for each pixel is uniformly optimized by utilizing the fact that the quantization noise power increases in proportion to the square of the quantization step width, and is the same as in the first embodiment. It can also comprise so that the effect of this may be acquired.
[0103]
That is, in the encoding device, the quantization step width of the quantizer 14 is corrected by scalar multiplication by a value inversely proportional to the square root of the number of pixels to be encoded M in the RS-DCT unit 10, and further the inverse in the decoding device. An equivalent effect can also be obtained by correcting the quantization step width of the quantizer 22 by scalar multiplication with a value proportional to the square root of the number of pixels to be encoded M.
[0104]
However, since the quantization processing is division opposite to multiplication, the quantization step width in the encoding device is the same as that in the first embodiment.
[0105]
[Outside 13]

[0106]
It is necessary to multiply the scalar using.
[0107]
【The invention's effect】
As described above, according to the present invention, boundary information of an image composed of a plurality of components. Block area information that is Encoding process based on Out When the encoded data output from the encoded apparatus is decoded by the decoding apparatus, a decoding process is performed so as to minimize the total distortion amount of the decoded image. Apply The effect that can be obtained.
[Brief description of the drawings]
FIG. 1 is a block diagram illustrating a configuration of a first embodiment of an image encoding device to which the present invention has been applied.
FIG. 2 is a block diagram illustrating a configuration of a first embodiment of an image decoding device to which the present invention has been applied.
[Explanation of symbols]
1, 2 block circuit
3 DCT base supply circuit
4,5 Support area correction circuit
6a, 6b, 6c Coefficient derivation circuit
7 Judgment circuit
8 Switching circuit
9 Combinational circuit
10 RS-DCT section
12,24 multiplier
14 Quantizer
16 Variable length encoder
20 Variable length decoder
22 Inverse quantizer
26 Two-dimensional inverse DCT section
28 Selector
29 Mask pattern

Claims (6)

An image composed of a plurality of constituent elements is divided into a plurality of blocks having a predetermined size composed of a predetermined number of pixels, and a pixel signal vector corresponding to at least one image region composed of a single constituent element included in the block is obtained. Conversion means for converting and outputting as a weighted linear sum of orthogonal transform basis vectors derived for each of the at least one image area based on block area information which is boundary information of the component elements, and quantizing the signal from the conversion means Quantizing means for converting the orthogonal transform basis vector for each of the encoding target image regions specified based on the block region information for a predetermined component corresponding to the encoding target image region of the discrete cosine transform base vector. held masked process to 0 the component other than the predetermined components, and it is divided by the square of the norm of the masked vector An image coding device is an orthogonal transform basis vector newly derived for each of the by scalar multiplication least one image area by,
The weighted transform coefficient from the transform unit is scalar multiplied by a value obtained by dividing the square root of the number of pixels by the number of pixels of the block for each of the encoding target image regions included in the block, and the input of the quantization unit. An image encoding apparatus characterized by: An image composed of a plurality of constituent elements is divided into a plurality of blocks having a predetermined size composed of a predetermined number of pixels, and a pixel signal vector corresponding to at least one image region composed of a single constituent element included in the block is obtained. Conversion means for converting and outputting as a weighted linear sum of orthogonal transform basis vectors derived for each of the at least one image area based on block area information which is boundary information of the component elements, and quantizing the signal from the conversion means Quantizing means for converting the orthogonal transform basis vector for each of the encoding target image regions specified based on the block region information for a predetermined component corresponding to the encoding target image region of the discrete cosine transform base vector. held masked process to 0 the component other than the predetermined components, and it is divided by the square of the norm of the masked vector An image coding device is an orthogonal transform basis vector newly derived for each of the by scalar multiplication least one image area by,
The signal level of the pixel signal vector is scalar multiplied by the value obtained by dividing the square root of the number of pixels by the number of pixels of the block for each of the encoding target image areas included in the block, and used as the input of the conversion unit. An image encoding device characterized by the above. An image composed of a plurality of constituent elements is divided into a plurality of blocks having a predetermined size composed of a predetermined number of pixels, and a pixel signal vector corresponding to at least one image region composed of a single constituent element included in the block is obtained. Conversion means for converting and outputting as a weighted linear sum of orthogonal transform basis vectors derived for each of the at least one image area based on block area information which is boundary information of the component elements, and quantizing the signal from the conversion means Quantizing means for converting the orthogonal transform basis vector for each of the encoding target image regions specified based on the block region information for a predetermined component corresponding to the encoding target image region of the discrete cosine transform base vector. held masked process to 0 the component other than the predetermined components, and it is divided by the square of the norm of the masked vector An image coding device is an orthogonal transform basis vector newly derived for each of the by scalar multiplication least one image area by,
The quantization means scalar-multiplies the quantization step width by a value obtained by dividing the number of pixels of the block by the square root of the number of pixels for each of the encoding target image regions included in the block. Device. 2. A pixel signal vector corresponding to at least one image area composed of a single component included in a block composed of a predetermined number of pixels and divided by a predetermined size, which is output from the image encoding device according to claim 1 Inverse quantization means for inversely quantizing encoded data obtained by encoding a weighted coefficient, and orthogonal transform derived for each of the at least one image area based on block area information that is boundary information of the component Inverse transformation means for decoding a pixel signal vector corresponding to the at least one image area by performing weighted linear addition of the basis vectors, and the orthogonal transformation basis vector is included in an encoding target image area of the discrete cosine transform basis vector. A component other than the predetermined component that holds the corresponding predetermined component for each of the encoding target image regions specified based on the block region information. 0 and the masking processing on the, and a image decoding apparatus is an orthogonal transform basis vector newly derived for each of the by scalar multiplication one or more image regions divided by the square of the norm of the masked vector And
The weighting coefficient obtained by the inverse quantization is scalarized by a value obtained by dividing the number of pixels of the block by the square root of the number of pixels for each encoding target image area included in the block, and weighting the orthogonal transform base vector An image decoding apparatus characterized by performing linear addition. A pixel signal vector corresponding to at least one image area composed of a single component included in a block composed of a predetermined number of pixels and divided by a predetermined size, which is output from the image encoding device according to claim 2 Inverse quantization means for inversely quantizing encoded data obtained by encoding a weighted coefficient, and orthogonal transform derived for each of the at least one image area based on block area information that is boundary information of the component Inverse transformation means for decoding a pixel signal vector corresponding to the at least one image area by performing weighted linear addition of the basis vectors, and the orthogonal transformation basis vector is included in an encoding target image area of the discrete cosine transform basis vector. A component other than the predetermined component that holds the corresponding predetermined component for each of the encoding target image regions specified based on the block region information. 0 and the masking processing on the, and a image decoding apparatus is an orthogonal transform basis vector newly derived for each of the by scalar multiplication one or more image regions divided by the square of the norm of the masked vector And
A pixel corresponding to the at least one image region is obtained by scalar-multiplying the inverse transform means output by a value obtained by dividing the number of pixels of the block by the square root of the number of pixels for each encoding target image region included in the block. An image decoding apparatus for decoding a signal vector. 4. A pixel signal vector corresponding to at least one image area composed of a single component included in a block composed of a predetermined number of pixels and divided by a predetermined size, which is output from the image encoding device according to claim 3. Inverse quantization means for inversely quantizing encoded data obtained by encoding a weighted coefficient, and orthogonal transform derived for each of the at least one image area based on block area information that is boundary information of the component Inverse transformation means for decoding a pixel signal vector corresponding to the at least one image area by performing weighted linear addition of the basis vectors, and the orthogonal transformation basis vector is included in an encoding target image area of the discrete cosine transform basis vector. A component other than the predetermined component that holds the corresponding predetermined component for each of the encoding target image regions specified based on the block region information. 0 and the masking processing on the, and a image decoding apparatus is an orthogonal transform basis vector newly derived for each of the by scalar multiplication one or more image regions divided by the square of the norm of the masked vector And
The inverse quantization unit scalar-multiplies a quantization step width by a value obtained by dividing the square root of the number of pixels by the number of pixels of the block for each of the encoding target image regions included in the block. Decryption device.

Images