JP3777600B2 - Image information conversion device, coefficient calculation device, representative pixel signal setting device and method, storage device, recording medium, and program - Google Patents

Description

【０１０５】
式（４）のｅ＾2（ｅの2乗）を最小とする各予測係数ｗiは、ｅ＾2を予測係数ｗi（i=1,2・・・・）で偏微分した式（５）に変形し、ｉの各値について偏微分値が０となるものとして計算される。
【０１０６】
【数２】

【０１０７】
式（５）から各予測係数ｗiを定める具体的な手順について説明する。式（６），式（７）のようにＸji，Ｙiを定義するとき、式（５）は、式（８）の行列式の形に変形される。
【０１０８】
【数３】

【０１０９】
【数４】

【０１１０】
【数５】

【０１１１】
この式（８）が、一般に正規方程式と呼ばれるものである。ここで、Ｘji（ｊ，ｉ＝１，２・・・・ｎ），およびＹi （ｉ＝１，２・・・・ｎ）は、教師画像、および、生徒画像に基づいて計算される。すなわち、正規方程式生成部１０６は、ＸjiおよびＹiの値を算出して正規方程式からなる式（８）を決定する。さらに、係数決定部１０８は、式（８）を解いて各予測係数ｗiを決定する。
【０１１２】
次に、図２０のフローチャートを参照して、図１９の学習装置が予測係数を学習する処理について説明する。
【０１１３】
ステップＳ７１において、教師画像としてのコンポーネント信号がNTSCエンコーダ１０１、および、正規方程式生成部１０６に入力される。ステップＳ７２において、NTSCエンコーダ１０１は、入力されたコンポーネント信号よりNTSC方式のコンポジット信号からなる生徒画像を生成してクラスタップ抽出部１０２、および、予測タップ抽出部１０７に出力する。
【０１１４】
ステップＳ７３において、クラスタップ抽出部１０２は、クラスタップを抽出してベクトル生成部１０３に出力する。尚、この処理は、図１０のフローチャートのステップＳ４の処理と同じものである。
【０１１５】
ステップＳ７４において、ベクトル生成部５２は、クラスタップ抽出部１０２により抽出されたクラスタップの画素値を読出し、これらの値からなるベクトルを生成して（ベクトル量子化して）、クラスコード決定部１０４に出力する。尚、この処理は、図１０のフローチャートのステップＳ５の処理と同じものである。
【０１１６】
ステップＳ７５において、クラスコード決定部１０４は、入力されたベクトルと、予め代表ベクトルメモリ１０５に記憶された代表ベクトル群と比較し、最も近い代表ベクトルを決定し、対応するクラス（インデックス）コードを予測演算部１０６に出力する。
【０１１７】
ステップＳ７６において、予測タップ抽出部５５は、生徒画像の予測タップを抽出して正規方程式生成部１０６に出力する。
【０１１８】
ステップＳ７７において、正規方程式生成部１０６は、クラスコード決定部１０４より入力されたクラスコード、予測タップ抽出部１０７より入力された予測タップ、および、教師画像として入力されたコンポーネント信号から上述の式（８）で示した正規方程式を生成し、クラスコード決定部１０４より入力されたクラスコードと共に係数決定部１０８に出力する。
【０１１９】
ステップＳ７８において、係数決定部１０８は、上述の式（８）からなる正規方程式を解いて、予測係数を決定し、クラスコードに対応付けて出力し、係数メモリ５７に記憶させる。
【０１２０】
ステップＳ７９において、全ての画素について処理が施されたか否かが判定され、全ての画素について処理が施されていないと判定された場合、その処理は、ステップＳ７３に戻る。すなわち、全ての画素の処理が終了されるまで、ステップＳ７３乃至Ｓ７９の処理が繰り返される。ステップＳ７９において、全ての画素について処理が施されたと判定された場合、その処理は、終了する。
【０１２１】
以上の処理により、着目画素とその周辺の位相に制限されないクラスタップがベクトル量子化されたベクトルを使用して、クラスを決定することにより適正なクラス分類がなされ、予測係数が演算されるので、コンポジット信号からコンポーネント信号の変換処理により生じる画質劣化を抑制できる予測係数を求めることが可能となる。また、上述においては、輝度信号Ｙを求める予測係数を設定する例について説明してきたが、色差信号Ｒ−Ｙ，Ｂ−Ｙについても同様にして求めることができる。
【０１２２】
次に、図２１乃至図２８を参照して、従来の画像変換装置と本発明を適用したテレビジョン受像機とのコンポジット信号からコンポーネント信号への変換時のSNR（Signal to Noise：入力されたコンポジット信号に対応する真のコンポーネント信号と、変換されたコンポーネント信号との比率）の比較結果について説明する。
【０１２３】
図２１，図２２においては、図中、横方向はフレーム数を示し、縦方向はSNR（ｄB）であり、本発明を適応したテレビジョン受像機のSNRが実線で、従来の画像変換装置のSNRが点線でそれぞれ示されている。また、本発明を適応したテレビジョン受像機においては、図２３で示すように、着目画素がフィールド＃０上に存在し、クラスタップが着目画素を中心とした水平方向の５画素、着目画素の上下の画素を中心とした水平方向の５画素からなる計１５タップ（クラス数＝３２）で、かつ、予測タップは図４で示した１２タップを使用した場合であり、また、従来の画像変換装置においては、着目画素がフィールド＃０上に存在するとき、クラスタップは、着目画素が、図３で示した５画素（クラス数＝３１）で、かつ、予測タップが図４で示した１２タップの場合を示している。
【０１２４】
さらに、図２１では、静止画像領域を多く含む画像（704画素×484画素，６０フレーム，Ｙ＋Ｑ）の、図２２においては、動画像領域を多く含む画像（704画素×484画素，６０フレーム，Ｙ＋Ｑ）の、それぞれの輝度信号Ｙについての結果である。
【０１２５】
図２１において、従来の画像変換装置の場合、SNRは37.5dB付近であるのに対して、本発明を適用したテレビジョン受像機の場合、SNRは38ｄB付近であり、本発明を適用することにより、約0.5ｄB程のSNRが改善されていることが示されている。
【０１２６】
また、図２２については、従来の画像変換装置の場合と本発明を適用したテレビジョン受像機の場合とでは、全体として１乃至３dB程度本発明を適用した場合の方が、従来の場合のSNRより改善されていることが示されている。
【０１２７】
また、図２４，図２５においても、図２１，図２２同様に、図中、横方向はフレーム数を示し、縦方向はSNRであり、本発明を適応したテレビジョン受像機のSNRが実線で、また、従来の画像変換装置のSNRが点線で示されている。このとき、従来の画像変換装置において、クラスタップは、図２６で示されているように、図３で示したクラスタップに、２フィールド前のフィールド＃−２上の着目画素と同位相の４画素を加えた合計９タップ（クラス数＝５１１）を使用し、予測タップは、図２７で示すように、図４で示した１２タップに、その直前のフィールド＃−１上の着目画素位置の上下に存在する画素を中心とした左右の画素からなる６画素と、さらに、その前のフィールド＃−２上の着目画素位置の画素とその左右の画素、並びに、着目画素の上下の位置に存在する画素と、その左右の画素からなる９画素を加えた合計２７タップ（オフセットタップを含む）を使用している。
【０１２８】
さらに、本発明を適応したテレビジョン受像機においては、クラスタップは、図２８で示すように、図２３で示したクラスタップに、直前のフィールド＃−１上の着目画素位置の上下の画素とその左右の画素からなる６画素と、さらに、その前のフィールド＃−２の着目画素とその左右の画素、並びに、着目画素の上下の画素と、それぞれの左右の画素の9画素を加えた合計３０タップ（クラス数＝５１２）とし、予測タップは、従来の画像変換装置と同様のものを使用した（図２７）。
【０１２９】
また、図２４は、静止画像領域を多く含む画像（704画素×484画素，６０フレーム，Ｙ＋Ｑ）の、図２５は、動画像領域を多く含む画像（704画素×484画素，６０フレーム，Ｙ＋Ｑ）の、それぞれの輝度信号Ｙについての結果である。
【０１３０】
図２４においては、従来の画像変換装置のSNRについては、43ｄB付近であるのに対して、本発明を適用したテレビジョン受像機のSNRは43.5ｄB付近であり、従来の画像変換装置に比べて、約0.5dB程度のSNRが向上されている。また、図２５においては、全体として0.5dB程度のSNRの向上が見られる。
【０１３１】
尚、図示しないが、色差信号、および、その他位相の信号についても図２１，図２２，図２４，図２５と同様の改善がなされることが確認されている。
【０１３２】
図２１乃至図２８を参照して説明したように、クラスタップ数や、静止画像領域の多い画像であるか、または、動画像領域の多い画像であるかに関わらず、代表ベクトルを用いたクラス分類適応処理は、コンポジット信号からコンポーネント信号への変換時のSNRを効果的に改善されており、結果として、画質劣化が抑制されていることが示されている。
【０１３３】
また、上述のベクトルは、各要素を正規化したものであってもよい。すなわち、例えば、（Ｘ１００，Ｘ１０１，Ｘ１０２・・・Ｘ２００）といったようなベクトルが存在する場合、これらの各要素の最大値Max（Ｘ１００，Ｘ１０１，Ｘ１０２・・・Ｘ２００）を求めて、各要素を除することにより、（Ｘ１００／Max（Ｘ１００，Ｘ１０１，Ｘ１０２・・・Ｘ２００），Ｘ１０１／Max（Ｘ１００，Ｘ１０１，Ｘ１０２・・・Ｘ２００），Ｘ１０２／Max（Ｘ１００，Ｘ１０１，Ｘ１０２・・・Ｘ２００），・・・Ｘ２００／Max（Ｘ１００，Ｘ１０１，Ｘ１０２・・・Ｘ２００））といったように正規化して表現するようにしても良い。
【０１３４】
このように正規化したベクトルを使用した場合、図９で示したインデックスは、図２９で示すように表現されることになる。
【０１３５】
尚、以上の例においては、コンポジット映像信号からコンポーネント映像信号（Ｙ，Ｃｒ，Ｃｂ）を生成するようにしたが、例えば、図３０で示すように、クラス分類適応処理回路１２１により、VIF回路３３の出力するコンポジット映像信号から、原色ＲＧＢ信号を直接生成するようにすることもできる。この場合においても、クラス分類適応処理回路１２１の構成は、クラス分類適応処理回路３５と同様であるが、係数メモリ５７に記憶される予測係数を決定する学習処理における生徒画像は、コンポジット信号であるが、教師画像は、原色ＲＧＢ信号となり、決定される予測係数も原色ＲＧＢ信号に対応したものとなる。ただし、代表ベクトルメモリ５４に記憶された代表ベクトルはそのままでよい。
【０１３６】
さらに、以上においては、NTSC方式のコンポジット信号を用いた例について説明してきたが、例えば、PAL（Phase Alternating Line）方式のコンポジット信号からコンポーネント信号や原色ＲＧＢ信号に変換する場合においても、同様の構成で実現させることができる。
【０１３７】
以上によれば、代表ベクトルによりクラスコードを決定するようにしたので、コンポジット信号の全ての位相についての相対的な関係を考慮した適正な着目画素のクラス分類を行うことができ、コンポジット信号からコンポーネント信号への変換時に生じる画質劣化を抑制させることが可能となる。
【０１３８】
上述した一連の処理は、ハードウェアにより実行させることもできるが、ソフトウェアにより実行させることもできる。一連の処理をソフトウェアにより実行させる場合には、そのソフトウェアを構成するプログラムが、専用のハードウェアに組み込まれているコンピュータ、または、各種のプログラムをインストールすることで、各種の機能を実行させることが可能な、例えば汎用のパーソナルコンピュータなどに記録媒体からインストールされる。
【０１３９】
図３１は、テレビジョン受像機、代表ベクトル設定装置、または、学習装置をそれぞれソフトウェアにより実現する場合のパーソナルコンピュータの一実施の形態の構成を示している。パーソナルコンピュータのCPU２０１は、パーソナルコンピュータの動作の全体を制御する。また、CPU２０１は、バス２０４および入出力インタフェース２０５を介してユーザからキーボードやマウスなどからなる入力部２０６から指令が入力されると、それに対応してROM(Read Only Memory)２０２に格納されているプログラムを実行する。あるいはまた、CPU２０１は、ドライブ２１０に接続された磁気ディスク２１１、光ディスク２１２、光磁気ディスク２１３、または半導体メモリ２１４から読み出され、記憶部２０８にインストールされたプログラムを、RAM(Random Access Memory)２０３にロードして実行する。これにより、上述した画像処理装置の機能が、ソフトウェアにより実現されている。さらに、CPU２０１は、通信部２０９を制御して、外部と通信し、データの授受を実行する。
【０１４０】
プログラムが記録されている記録媒体は、図３１に示すように、コンピュータとは別に、ユーザにプログラムを提供するために配布される、プログラムが記録されている磁気ディスク２１１（フレキシブルディスクを含む）、光ディスク２１２（CD-ROM(Compact Disc-Read Only Memory)，DVD（Digital Versatile Disc）を含む）、光磁気ディスク２１３（MD（Mini-Disc）を含む）、もしくは半導体メモリ２１４などよりなるパッケージメディアにより構成されるだけでなく、コンピュータに予め組み込まれた状態でユーザに提供される、プログラムが記録されているROM２０２や、記憶部２０８に含まれるハードディスクなどで構成される。
【０１４１】
尚、本明細書において、記録媒体に記録されるプログラムを記述するステップは、記載された順序に沿って時系列的に行われる処理は、もちろん、必ずしも時系列的に処理されなくとも、並列的あるいは個別に実行される処理を含むものである。
【０１４２】
【発明の効果】
本発明の画像情報変換装置および方法、並びに第１のプログラムによれば、着目画素をクラスに分類するための複数の第１の画素信号を入力コンポジット信号から抽出し、入力コンポジット信号の各画素をクラスに分類するための複数の画素信号のパターンをベクトル量子化し、それを用いたベクトル空間上での統計的学習により予めクラス毎に定められた代表パターンを記憶し、抽出した複数の第1の画素信号のパターンに、最も近い代表パターンを検索し、検索した代表パターンが定められているクラスを、着目画素のクラスに分類し、入力コンポジット信号から着目画素を変換するための複数の第２の画素信号を抽出し、第２の画素信号と、分類したクラス毎に設定されている係数との積和演算により、着目画素の画素信号を変換するようにした。
【０１４４】
本発明の係数算出装置および方法、並びに第２のプログラムによれば、入力された画像信号から入力コンポジット信号を生成し、着目画素をクラスに分類するための複数の第１の画素の画素信号を入力コンポジット信号から抽出し、入力コンポジット信号の各画素をクラスに分類するための複数の画素信号のパターンをベクトル量子化し、それを用いたベクトル空間上での統計的学習により予めクラス毎に定められた代表パターンを記憶し、抽出した複数の第1の画素信号のパターンに、最も近い代表パターンを検索し、検索した代表パターンが定められている前記クラスを、着目画素のクラスに分類し、入力コンポジット信号から複数の第２の画素信号を抽出し、第２の画素信号と入力された画像信号とを用いた正規方程式より、分類したクラス毎に係数を算出するようにした。
【０１４５】
本発明の代表パターン設定装置および方法、並びに第３のプログラムによれば、入力コンポジット信号の各画素をクラスに分類するための複数の画素信号を抽出し、複数の画素信号をベクトル量子化し、ベクトル量子化した複数の画素信号から、ベクトル空間上での代表ベクトルをクラス数に応じて計算し、計算した代表ベクトルを代表パターンとしてクラスに対応付けて設定するようにした。
【０１４６】
結果として、いずれにおいても、コンポジット信号の全ての位相についての相対的な関係を考慮した適正な着目画素のクラス分類を行うことができ、コンポジット信号からコンポーネント信号への変換時に生じる画質劣化を抑制させることが可能となる。
【図面の簡単な説明】
【図１】従来の画像変換装置の構成例を示すブロック図である。
【図２】 NTSCコンポジット信号を説明する図である。
【図３】クラスタップを説明する図である。
【図４】予測タップを説明する図である。
【図５】従来の学習装置の構成例を示すブロック図である。
【図６】本発明を適用したテレビジョン受像機の一実施の形態の構成を示すブロック図である。
【図７】図６のクラス分類適応処理回路を説明する図である。
【図８】図６のクラス分類適応処理回路の構成を説明するブロック図である。
【図９】図８の代表ベクトルメモリに記憶される代表ベクトルを説明する図である。
【図１０】図６のテレビジョン受像機の表示処理を説明するフローチャートである。
【図１１】代表ベクトルを選択する処理を説明する図である。
【図１２】代表ベクトル設定装置と学習装置を説明する図である。
【図１３】代表ベクトル設定装置の構成を説明するブロック図である。
【図１４】図１３の代表ベクトル設定装置の代表ベクトル設定処理を説明するフローチャートである。
【図１５】図１３の代表ベクトル計算部による代表ベクトル決定処理を説明するフローチャートである。
【図１６】代表ベクトルの計算方法を説明する図である。
【図１７】代表ベクトルの計算方法を説明する図である。
【図１８】代表ベクトルの計算方法を説明する図である。
【図１９】学習装置を説明するブロック図である。
【図２０】学習装置の学習処理を説明するフローチャートである。
【図２１】従来の画像変換装置と本発明を適用したテレビジョン受像機のSNRの比較結果を示す図である。
【図２２】従来の画像変換装置と本発明を適用したテレビジョン受像機のSNRの比較結果を示す図である。
【図２３】図２１，図２２のときの本発明を適用したテレビジョン受像機のクラスタップを示す図である。
【図２４】従来の画像変換装置と本発明を適用したテレビジョン受像機のSNRの比較結果を示す図である。
【図２５】従来の画像変換装置と本発明を適用したテレビジョン受像機のSNRの比較結果を示す図である。
【図２６】図２４，図２５のときの従来の画像変換装置のクラスタップを示す図である。
【図２７】図２４，図２５のときの予測タップを示す図である。
【図２８】図２４，図２５のときの本発明を適用したテレビジョン受像機のクラスタップを示す図である。
【図２９】図９の代表ベクトルを正規化した場合の例を説明する図である。
【図３０】本発明を適用したテレビジョン受像機の他の構成を示すブロック図である。
【図３１】媒体を説明する図である。
【符号の説明】
３５ クラス分類適応処理回路， ５１ クラスタップ抽出部， ５２ ベクトル生成部， ５３ クラスコード決定部， ５４ 代表ベクトルメモリ， ５５ 予測タップ抽出部， ５６ 予測演算部， ５７ 係数メモリ， ６１ 代表ベクトル設定装置， ６２ 学習装置， ８１ 設定タップ抽出部， ８２ ベクトル生成部， ８３ ベクトル空間データ保持部， ８４ 代表ベクトル計算部， １０１ NTSCエンコーダ， １０２ クラスタップ抽出部， １０３ ベクトル生成部， １０４ クラスコード決定部， １０５ 代表ベクトルメモリ， １０６ 正規方程式生成部， １０７ 予測タップ抽出部， １０８ 係数決定部， １２１ クラス分類適応処理回路[0001]
BACKGROUND OF THE INVENTION
  The present inventionImage information conversion device, coefficient calculation device,Regarding representative pixel signal setting device and method, storage device, recording medium, and program, in particular, it is possible to suppress deterioration in image quality that occurs when converting a composite signal to a component signal.Image information conversion device, coefficient calculation device,The present invention relates to a representative pixel signal setting device and method, a storage device, a recording medium, and a program.
[0002]
[Prior art]
Generally, in a television receiver, a composite signal of a video signal is converted into a component signal and displayed as a video. As video signals, NTSC (National Television Standards Committee) type composite video signals are widely used.
[0003]
Conventionally, a composite video signal is separated into a luminance signal Y and a chroma signal C, and then the chroma signal C is demodulated into a component signal composed of a baseband luminance signal Y and color difference signals RY and BY. It was converted. However, the circuit required for these processes not only has a complicated structure, but also has a large scale and a high cost. Therefore, a method using a class classification adaptive process has recently been proposed. .
[0004]
In this method using the class classification adaptive processing, in order to obtain component signals such as the luminance signal Y and the color difference signals RY and BY at the target pixel, first, they are arranged in the vicinity including the target pixel. A feature quantity of a pixel (this pixel is called a class tap) is obtained from the composite signal, and a class classified in advance for each feature quantity is specified. Then, a calculation process using a coefficient set in advance for each class in a pixel including a target pixel and a pixel composed of a composite signal in the vicinity thereof (this pixel is called a prediction tap and may be the same as a class tap). To obtain the component signal of the pixel of interest.
[0005]
FIG. 1 shows the configuration of an image conversion apparatus when the class classification adaptation process is used. The class classification unit 11 extracts a class tap corresponding to the target pixel from the input composite signal, compares the pixel value of the pixel serving as the class tap with a predetermined calculation, and compares the value with a predetermined threshold value. 1 or 0 is set by, and binarized information is generated from the number of class taps. Further, the class classification unit 11 obtains the pattern of the binarized information as the feature amount of the target pixel, determines the class of the target pixel according to the feature amount, and outputs the determined class data to the prediction unit 12. The prediction unit 12 extracts a corresponding prediction tap, and based on the class data input from the class classification unit 11, performs processing using a prediction coefficient input in advance and stored in a built-in memory to generate a component signal And output.
[0006]
Next, the operation of the image conversion apparatus will be described. The class classification unit 11 extracts class taps from the composite signal. For example, each pixel composed of a composite signal is a four-phase signal based on a combination of a luminance signal Y, an I signal, and a Q signal as shown in FIG. That is, as shown in FIG. 2A, each pixel on each field is indicated by a white circle (Y + I), a white square (Y + Q), a black circle (Y-I), and a black square ( The four phases of YQ) are sequentially arranged in the horizontal direction, and 180 phases are inverted one by one in the vertical direction. Here, in FIG. 2A, # 0 to # 3 indicate field numbers, which are arranged in time series. Also, the crosses in the figure indicate the target pixel position.
[0007]
The phase of the pixels on each field changes with time. For example, the pixels in the columns L0 to L3 including the target pixel position change periodically as shown in FIG. 2B. In other words, the phase of each pixel is inverted by 180 degrees every two fields in the vertical direction as time advances.
[0008]
When the target pixel is present on the field # 0 in the composite signal as illustrated in FIG. 2, for example, as illustrated in FIG. 3, the class classification unit 11 centered on the target pixel on the field # 0. Five pixels (including the pixel of interest) composed of four pixels having the same phase and existing at close positions are extracted as class taps.
[0009]
Further, the class classification unit 11 converts a value obtained by processing the pixel value of each pixel of this class tap by a predetermined calculation from a binarization information pattern (in this example, 5 for each class tap from the magnitude relationship with a predetermined threshold). Bit pattern) is generated, a class set in advance for each pattern is determined, and is output to the prediction unit 12 as class data.
[0010]
For example, the prediction unit 12 extracts a prediction tap as illustrated in FIG. 4, reads each pixel value of the prediction tap, performs arithmetic processing using a prediction coefficient stored in advance in a built-in memory, and performs component processing. Is generated and output. In FIG. 4, the prediction tap includes five pixels in the horizontal direction centered on the pixel of interest on field # 0, 11 pixels above and below the pixel of interest, and 11 pixels on the left and right thereof. A total of 12 taps of offset taps (offset) for correcting the deviation. The prediction unit 12 performs a calculation process using a prediction coefficient corresponding to the class data on each pixel value of the prediction tap including the 12 pixels to generate a component signal.
[0011]
A prediction coefficient is produced | generated by the learning process by a learning apparatus as shown in FIG. In this learning process, a prediction coefficient is obtained using a relationship between a component signal called a teacher image and a composite signal called a student image. In other words, this is a process of calculating a prediction coefficient by calculation from the mutual relationship between a teacher image as a target value of an output result and a student image that is an input image signal.
[0012]
The class classification unit 21 of the learning device is the same as the class classification unit 11 of the image conversion apparatus shown in FIG. 1, and a class for determining the class of the pixel of interest from the composite signal as the input student signal. A tap is extracted, and the pixel value of each pixel is compared with a value processed by a predetermined calculation with a predetermined threshold, and a pattern of binarized information is generated from the magnitude relationship in the same manner as the class classification unit 11 described above. The class of the target pixel is determined from the pattern, and class data is output to the prediction coefficient calculation unit 22. The prediction coefficient calculation unit 22 extracts a prediction tap from the composite signal, calculates a prediction coefficient from the relationship with the component signal that is a teacher image, and outputs it together with the class data.
[0013]
Next, the learning process of the learning device will be described.
[0014]
The class classification unit 21 extracts a class tap from the input composite signal as a student image, compares each pixel value of the class tap with a predetermined threshold value, and generates binarized information from the magnitude relationship. Thereafter, the class classification unit 21 determines the class of the pixel of interest from the pattern and outputs it to the prediction coefficient calculation unit 22.
[0015]
That is, for example, when the target pixel exists on the field # 0 in FIG. 2, the class classification unit 21 is the same as the target pixel shown in FIG. A total of 5 pixels of 4 pixels in phase are extracted as class taps. Further, the class classification unit 21 compares a value obtained by processing each pixel value of the extracted class tap by a predetermined calculation with a predetermined threshold value, and generates a pattern of binarized information from the magnitude relationship. The class classification unit 21 determines the class of the pixel of interest based on the pattern and outputs it to the prediction coefficient calculation unit 22.
[0016]
The prediction coefficient calculation unit 22 extracts a prediction tap corresponding to the target pixel from the composite signal that is a student image, and calculates and outputs a prediction coefficient from the relationship with the component signal as the teacher image. For example, as shown in FIG. 4, the prediction tap is composed of 5 pixels in the horizontal direction centering on the pixel of interest on the field # 0, pixels above and below the pixel of interest, and a total of 11 pixels on each of the left and right pixels. In addition, there are 12 offset taps.
[0017]
In this manner, the prediction coefficient calculated in advance by the learning process of the learning device is stored in the prediction unit 11 of the image conversion device. Thereafter, the image conversion apparatus converts the composite signal into a component signal using the input prediction coefficient.
[0018]
[Problems to be solved by the invention]
However, in the class tap selection method as described above, the feature amount of the pixel of interest can be obtained only with pixel values in the same phase, and the feature amount including pixels in different surrounding phases is obtained and the corresponding class is obtained. Cannot be set. That is, as shown in FIG. 2, each pixel has a different phase according to time and according to a position in the horizontal and vertical directions, so that pixel values between different phases cannot be compared and are different. The class determination process for the pixel of interest cannot be performed in consideration of the relationship between pixels having a phase. For this reason, the class of the pixel of interest may not be properly determined. As a result, image conversion processing by proper class classification processing cannot be performed, and there is a problem that image quality after conversion may be deteriorated. .
[0019]
The present invention has been made in view of such a situation, and has a simple and small circuit configuration, and further, class classification that appropriately reflects the relationship between the pixel of interest and its surrounding pixels. By doing so, it is possible to suppress deterioration in image quality that occurs during conversion from a composite signal to a component signal, particularly image quality deterioration that occurs when the class classification is not appropriate.
[0020]
[Means for Solving the Problems]
  The image information conversion apparatus according to the present invention includes a pixel of interest.To classifyFirst extraction means for extracting a plurality of first pixel signals from the input composite signal;Vector quantization of the pattern of multiple pixel signals for classifying each pixel of the input composite signal into a class, and using that in the vector spacePre-statistical learningPer classRepresentative pattern storage means for storing a predetermined representative pattern, representative pattern search means for searching for a representative pattern closest to the plurality of first pixel signal patterns extracted by the first extraction means, and representative pattern search Representative pattern retrieved by meansThe class where,Classify into the pixel class of interestFrom class classification means and input composite signalFor converting the pixel of interestA second extraction means for extracting a plurality of second pixel signals, a second pixel signal, and a class classified by the class classification means;Set for eachWith coefficientSum of productsAnd pixel-of-interest signal conversion means for converting a pixel signal of the pixel of interest by calculation.
[0021]
  Vector quantization means for vector quantization of the plurality of first pixel signals may be further provided, and the representative pattern storage means may include a composite signal.Classify the pixel of interest into a classOf the plurality of first pixel signals, a representative vector corresponding to a representative pattern composed of a plurality of representative first pixel signals can be stored.
[0022]
  The representative pattern search means includes a vector representative of the closest representative vector in the vector space and a vector quantized version of the plurality of first pixel signals extracted by the first extraction means.AsThe representative pattern can be searched as the closest representative pattern to the patterns of the plurality of first pixel signals extracted by the first extracting means.
[0023]
  The pixel-of-interest signal conversion means includes a second pixel signal and a class classified by the class classification means.Is setWith coefficientSum of productsBy calculation, the pixel signal of the pixel of interest can be converted into a component signal.
[0024]
  The pixel-of-interest signal conversion means includes a second pixel signal and a class classified by the class classification means.Is setWith coefficientSum of productsBy calculation, the pixel signal of the pixel of interest can be converted into a luminance signal or a chroma signal.
Display means for displaying the pixel signal of the pixel of interest converted into the component signal can be further provided.
[0025]
  The image information conversion method of the present invention includes a pixel of interestTo classifyA first extraction step of extracting a plurality of first pixel signals from the input composite signal;Vector quantization of the pattern of multiple pixel signals for classifying each pixel of the input composite signal into a class, and using that in the vector spacePre-statistical learningPer classA representative pattern storing step for storing a predetermined representative pattern, a representative pattern searching step for searching for a representative pattern closest to the plurality of first pixel signal patterns extracted in the processing of the first extracting step, and a representative Representative pattern searched by pattern search step processingSaid class is defined,Classify into the pixel class of interestFrom classification step and composite signalFor converting the pixel of interestThe pixel signal of the pixel of interest is converted by calculating a second extraction step for extracting a plurality of second pixel signals, the second pixel signal, and a coefficient corresponding to the class classified in the class classification step processing. And a target pixel signal conversion step.
[0026]
  The program of the first recording medium of the present invention is a pixel of interest.To classifyA first extraction control step for controlling extraction of a plurality of first pixel signals from an input composite signal;Vector quantization of the pattern of multiple pixel signals for classifying each pixel of the input composite signal into a class, and using that in the vector spacePre-statistical learningPer classControls the search of the representative pattern closest to the pattern of the plurality of first pixel signals whose extraction is controlled by the processing of the representative pattern storage control step for controlling the storage of the defined representative pattern and the first extraction control step. Representative pattern search control step to be performed and the representative pattern searched by the processing of the representative pattern search control stepSaid class is defined,Classification as pixel class of interestFrom the classification control step to control the input composite signalFor converting the pixel of interestA second extraction control step for controlling extraction of a plurality of second pixel signals, a second pixel signal, and a class whose classification is controlled by the processing of the class classification control stepSet for eachWith coefficientSum of productsAnd a pixel-of-interest signal conversion control step for controlling the conversion of the pixel signal of the pixel of interest by calculation.
[0027]
  The first program of the present invention is the pixel of interestTo classifyA first extraction control step for controlling extraction of a plurality of first pixel signals from an input composite signal;Vector quantization of the pattern of multiple pixel signals for classifying each pixel of the input composite signal into a class, and using that in the vector spaceThrough statistical learningFor each class in advanceControls the search of the representative pattern closest to the pattern of the plurality of first pixel signals whose extraction is controlled by the processing of the representative pattern storage control step for controlling the storage of the defined representative pattern and the first extraction control step. Representative pattern search control step to be performed and the representative pattern searched by the processing of the representative pattern search control stepSaid class is defined,Classification as pixel class of interestFrom the classification control step to control the input composite signalFor converting the pixel of interestA second extraction control step for controlling the extraction of a plurality of second pixel signals, a second pixel signal, and a class whose classification is controlled by the processing of the class classification control stepSet for eachWith coefficientSum of productsIt is characterized by causing a computer to execute a target pixel signal conversion control step for controlling conversion of a pixel signal of a target pixel by calculation.
[0032]
  A coefficient calculation apparatus according to the present invention includes a composite signal generation unit that generates an input composite signal from an input image signal,To classifyFirst extraction means for extracting pixel signals of a plurality of first pixels from an input composite signal;Vector quantization of the pattern of multiple pixel signals for classifying each pixel of the input composite signal into a class, and using that in the vector spacePre-statistical learningPer classRepresentative pattern storage means for storing a predetermined representative pattern, representative pattern search means for searching for a representative pattern closest to the plurality of first pixel signal patterns extracted by the first extraction means, and representative pattern search Representative pattern retrieved by meansSaid class is defined,Classify into the pixel class of interestFrom class classification means and input composite signalFor converting the pixel of interestA second extraction means for extracting a plurality of second pixel signals; and a second pixel signal and an input image signalFrom the normal equation using and, Classes classified by class classification meansEveryAnd a calculating means for calculating a coefficient.
[0033]
The storage device of the present invention stores the coefficient calculated by the coefficient calculation device according to claim 13.
[0034]
  The coefficient calculation method of the present invention includes a composite signal generation step of generating an input composite signal from an input image signal, and a pixel of interestTo classifyA first extraction step of extracting pixel signals of a plurality of first pixels from an input composite signal;Vector quantization of the pattern of multiple pixel signals for classifying each pixel of the input composite signal into a class, and using that in the vector spaceA representative pattern storage step for storing a representative pattern predetermined by statistical learning, and a representative pattern for searching for a representative pattern closest to the patterns of the plurality of first pixel signals extracted in the processing of the first extraction step Representative pixel signal searched by processing of search step and representative pattern search stepThe class where,Classify into the pixel class of interestFrom the classification step and the input composite signalFor converting the pixel of interestA second extraction step for extracting a plurality of second pixel signals; and a second pixel signal and an input image signalFrom the normal equation using andCategorized in the classification step processEveryAnd a calculating step for calculating a coefficient.
[0035]
  First of the present invention2The recording medium program includes a composite signal generation control step for controlling generation of an input composite signal from an input image signal, and a target pixel.To classifyA first extraction control step for controlling extraction of a pixel signal from an input composite signal of a plurality of first pixels;Vector quantization of the pattern of multiple pixel signals for classifying each pixel of the input composite signal into a class, and using that in the vector spacePre-statistical learningPer classControls the search of the representative pattern closest to the pattern of the plurality of first pixel signals whose extraction is controlled by the processing of the representative pattern storage control step for controlling the storage of the defined representative pattern and the first extraction control step. Representative pattern search control step to be performed and the representative pattern searched by the processing of the representative pattern search control stepThe class where,Classification as pixel class of interestFrom the classification control step to control the input composite signalFor converting the pixel of interestA second extraction control step for controlling the extraction of the plurality of second pixel signals; and the second pixel signal and the input image signalFrom the normal equation using andThe class whose classification is controlled by the processing of the class classification control stepEveryAnd a calculation control step for controlling calculation of the coefficient.
[0036]
  First of the present invention2The program includes a composite signal generation control step for controlling generation of an input composite signal from an input image signal, and a target pixel.To classifyA first extraction control step for controlling extraction of pixel signals of a plurality of first pixels from an input composite signal;Vector quantization of the pattern of multiple pixel signals for classifying each pixel of the input composite signal into a class, and using that in the vector spacePre-statistical learningPer classControls the search of the representative pattern closest to the pattern of the plurality of first pixel signals whose extraction is controlled by the processing of the representative pattern storage control step for controlling the storage of the defined representative pattern and the first extraction control step. Representative pattern search control step to be performed and the representative pattern searched by the processing of the representative pattern search control stepSaid class is defined,Classification as pixel class of interestClassification control step for controlling the second, second extraction control step for controlling the extraction of a plurality of second pixel signals from the input composite signal, and the image signal inputted with the second pixel signalFrom the normal equation using andThe class whose classification is controlled by the processing of the class classification control stepEveryA calculation control step for controlling calculation of a predetermined coefficient is caused to be executed by a computer.
[0037]
  The representative pattern setting device of the present invention provides each pixel of the input composite signal.To classifyExtraction means for extracting a plurality of pixel signals, vector quantization means for vector quantization of the plurality of pixel signals, and representative vectors in the vector space from the plurality of pixel signals vector quantized by the vector quantization means It is characterized by comprising calculation means for calculating according to the number of classes, and setting means for setting the representative vector calculated by the calculation means in association with the class as a representative pattern.
[0038]
The calculation means can be made to calculate a representative vector in the vector space according to the number of classes from a plurality of pixel signals vector quantized by the vector quantization means by the LBG algorithm.
[0039]
  The representative pattern setting method of the present invention provides each pixel of the input composite signal.To classifyA representative in vector space from an extraction step for extracting a plurality of pixel signals, a vector quantization step for vector quantization of the plurality of pixel signals, and a plurality of pixel signals vector quantized by the processing of the vector quantization step. It includes a calculation step for calculating a vector according to the number of classes, and a setting step for setting the representative vector calculated in the processing of the calculation step in association with a class as a representative pattern.
[0040]
  First of the present invention3The recording medium program for each pixel of the input composite signalTo classifyAn extraction control step for controlling extraction of a plurality of pixel signals, a vector quantization control step for controlling vector quantization of the plurality of pixel signals, and a plurality of pixels in which vector quantization is controlled by processing of the vector quantization control step From the signal, a calculation control step for controlling calculation according to the number of classes of the representative vector in the vector space, and a setting associated with the class as the representative pattern of the representative vector calculated in the processing of the calculation control step are controlled. And a setting control step.
[0041]
  First of the present invention3The program of each pixel of the input composite signalTo classifyAn extraction control step for controlling extraction of a plurality of pixel signals, a vector quantization control step for controlling vector quantization of the plurality of pixel signals, and a plurality of pixels in which vector quantization is controlled by processing of the vector quantization control step From the signal, a calculation control step for controlling calculation according to the number of classes of the representative vector in the vector space, and a setting associated with the class as the representative pattern of the representative vector calculated in the processing of the calculation control step are controlled. The setting control step is executed by a computer.
[0042]
  In the image information conversion apparatus and method and the first program of the present invention, the pixel of interestTo classifyA plurality of first pixel signals are extracted from the input composite signal;Vector quantization of the pattern of multiple pixel signals for classifying each pixel of the input composite signal into a class, and using that in the vector spacePre-statistical learningPer classA predetermined representative pattern is stored, and the closest representative pattern to the extracted plurality of first pixel signal patterns is searched, and the searched representative patternThe class for which,Classified into the pixel class of interestFrom the input composite signalFor converting the pixel of interestA plurality of second pixel signals are extracted, the second pixel signals, and the classified classesSet for eachWith coefficientSum of productsThe pixel signal of the pixel of interest is converted by the calculation.
[0044]
  Coefficient calculation apparatus and method of the present invention, and2In this program, the input composite signal is generated from the input image signal, and the pixel of interestTo classifyPixel signals of a plurality of first pixels are extracted from the input composite signal;Vector quantization of the pattern of multiple pixel signals for classifying each pixel of the input composite signal into a class, and using that in the vector spacePre-statistical learningPer classA predetermined representative pattern is stored, and the closest representative pattern to the extracted plurality of first pixel signal patterns is searched, and the searched representative patternThe class for which,Classify into pixel class of interestA plurality of second pixel signals are extracted from the input composite signal, and the second pixel signal and the input image signalFrom the normal equation using and, Classified classesEveryA coefficient is calculated.
[0045]
  Representative pattern setting apparatus and method of the present invention, and3In this program, each pixel of the input composite signalTo classifyA plurality of pixel signals are extracted, the plurality of pixel signals are vector-quantized, a representative vector in the vector space is calculated according to the number of classes from the plurality of vector-quantized pixel signals, and the calculated representative vector Are set in association with the class as a representative pattern.
[0046]
DETAILED DESCRIPTION OF THE INVENTION
FIG. 6 is a block diagram showing a configuration of an embodiment of a television receiver to which the present invention is applied.
[0047]
The tuner 32 demodulates the signal received by the antenna 31 and outputs it to the video intermediate frequency signal processing circuit (VIF circuit) 33. The VIF circuit 33 performs video intermediate frequency signal processing on the input signal and outputs the processed signal to an A / D (Analog / Digital) converter 34. The A / D converter 34 converts the NTSC composite video signal composed of the analog signal input from the VIF circuit 33 into a digital signal and outputs the digital signal to the class classification adaptive processing circuit 35. The class classification adaptive processing circuit 35 is a component composed of a luminance signal Y and color difference signals RY and BY signals from an NTSC composite video signal converted into a digital signal input from the A / D converter 34. The signal is directly generated by the classification adaptation process using the prediction coefficient.
[0048]
A prediction coefficient for predicting and generating a component signal from a composite video signal is generated by learning using a component signal as a teacher image and an NTSC composite video signal generated by NTSC modulation of the component signal as a student image. Then, a component signal is generated by performing mapping (prediction calculation processing) using the prediction coefficient. The matrix circuit 36 generates primary color RGB signals from the input luminance signal Y and the color difference signals RY and BY, and outputs the primary color RGB signals to the display device 37.
[0049]
More specifically, as shown in FIG. 7, the class classification adaptive processing circuit 35 vector-quantizes the pixel value of the class tap around the pixel of interest from the input NTSC composite signal, and any one of the representative vector groups generated in advance. Is selected from a representative vector group input in advance, and a prediction coefficient corresponding to a class of the pixel of interest set in advance is selected from the prediction coefficient group stored in advance and the prediction is performed. The composite signal is converted into a component signal by multiplying the prediction tap of the composite signal by the prediction coefficient using the coefficient. In FIG. 7, dotted arrows indicate data input in advance, and solid arrows indicate data processed in real time. Further, processing for generating the representative vector group and the prediction coefficient group will be described later.
[0050]
FIG. 8 shows a configuration example of the class classification adaptive processing circuit 35. The NTSC composite video signal input from the VIF circuit 33 is supplied to the class tap extraction unit 51 and the prediction tap extraction unit 55. The class tap extraction unit 51 extracts pixels (class taps) necessary for class classification from the composite video signal converted into the input digital signal, and outputs the extracted pixel (class tap) to the vector generation unit 52. The vector generation unit 52 generates a vector from each value of the input class tap. More specifically, when the pixel value of each pixel input as a class tap is, for example, (X100, X101... X200), the vector generation unit 52 uses these values as vector elements. Is generated and output to the class code determination unit 53.
[0051]
The class code determination unit 53 searches the representative vectors recorded in the representative vector memory 54 for the representative vector closest to the vector input from the vector generation unit 52, and determines the class code corresponding to the searched representative vector. And output to the prediction calculation unit 56. The representative vector memory 54 stores the number of representative vectors of the degree corresponding to the input vector corresponding to the class. For example, when the class tap is 2 pixels and the class is 11, 11 two-dimensional representative vectors are stored as shown in FIG. In FIG. 9, as a representative vector, an index for identifying a class is shown in the left column, and a representative vector corresponding to the index is shown on the right side. X1 to X11 and Y1 to Y11 indicate elements of each vector, that is, coordinate positions in the vector space.
[0052]
The representative vector is preset by the representative vector setting unit 61 (FIG. 13), and the processing will be described later.
[0053]
The prediction tap extraction unit 55 extracts a prediction tap necessary for predicting and generating a component signal from the composite video signal input from the VIF circuit 33 and outputs the prediction tap to the prediction calculation unit 56. Here, the prediction tap is a pixel that exists in a predetermined positional relationship with the target pixel to be processed.
[0054]
The prediction calculation unit 56 reads the prediction coefficient stored in advance in the coefficient memory 57 based on the class code input from the class code determination unit 53, and the prediction tap (each pixel extracted by the prediction tap extraction unit 55). Pixel value) is subjected to arithmetic processing using a prediction coefficient, and a component signal is generated and output.
[0055]
More specifically, the prediction calculation unit 56 extracts the prediction tap extracted from the pixel value at a predetermined pixel position of the NTSC composite video signal supplied from the prediction tap extraction unit 55, and the class code read from the coefficient memory 57. The component signal is obtained (predicted and estimated) by executing the product-sum operation shown in the following equation (1) using the prediction coefficient corresponding to.
[0056]
y = w1 * x1 + w2 * x2 ++ ... + wn * xn (1)
Here, y is a pixel of interest. In addition, x1,..., Xn are prediction taps, and w1,.
[0057]
In addition, although illustration is abbreviate | omitted, the circuit for producing | generating other color difference signal RY and color difference signal BY among component signals is also comprised similarly. The configuration is the same as that shown in FIG. 8, and the coefficient stored in the coefficient memory 57 is a prediction coefficient for generating the luminance signal Y in the case shown in FIG. When the signal RY or BY is generated, a prediction coefficient for generating the color difference signal RY or BY is stored.
[0058]
Further, the prediction coefficient learning process of the coefficient memory 57 will be described later with reference to FIG.
[0059]
Next, a process in which the television receiver shown in FIG. 6 converts a composite video signal into a component signal and displays it on the display device 37 will be described with reference to a flowchart of FIG.
[0060]
In step S <b> 1, the tuner 32 receives a signal of a predetermined frequency via the antenna 31, demodulates it, and outputs it to the VIF circuit 33. In step S <b> 2, the VIF circuit 33 generates an NTSC composite signal from the input signal and outputs the composite signal to the A / D converter 34. In step S <b> 3, the A / D converter 34 converts the input composite signal from an analog signal to a digital signal and outputs it to the class classification adaptive processing circuit 35.
[0061]
In step S <b> 4, the class tap extraction unit 51 extracts class taps and outputs them to the vector generation unit 52.
[0062]
In step S5, the vector generation unit 52 reads out the pixel values of the class tap extracted by the class tap extraction unit 51, generates a vector having these values as elements (vector quantization), and class code determination unit To 53.
[0063]
In step S6, the class code determination unit 53 compares the input vector with a representative vector group stored in advance in the representative vector memory 54, determines the closest representative vector, and selects the corresponding class (index) code. It outputs to the prediction calculation part 56. That is, for example, as shown in FIG. 9, when the class tap has two pixels and 11 representative vectors are set, when a vector corresponding to the class tap is input, the input vector is the most. The closest representative vector is selected. That is, as shown in FIG. 11, when the two-dimensional representative vector of FIG. 9 is indicated by rv1 to rv11 in the vector space (the vector number corresponds to the index number), it corresponds to the target pixel. When the obtained vector is indicated by v1, the representative vector rv11 closest to the vector v1 is selected. More specifically, the class code determination unit 53 obtains the distance between each representative vector and the input vector, and selects the representative vector that is the minimum value.
[0064]
In this case, the class code determination unit 53 outputs the class code of the representative vector rv11 shown in FIG. 11, that is, 11 corresponding to the index 11 as shown in FIG. 9, to the prediction calculation unit 56 as a class code.
[0065]
In step S <b> 7, the prediction calculation unit 56 reads out, from the coefficient memory 57, the prediction coefficient determined in advance by the learning process based on the class code input from the class code determination unit 53. This prediction coefficient is stored in the memory 57 in association with the class code. In step S <b> 8, the prediction tap extraction unit 55 extracts a prediction tap from the composite signal and outputs the prediction tap to the prediction calculation unit 56.
[0066]
In step S9, the prediction calculation unit 56 calculates the prediction tap input from the prediction tap extraction unit 55 using the prediction coefficient read from the coefficient memory 57, and generates a component signal.
[0067]
In step S10, it is determined whether or not signals have been converted for all pixels. If it is determined that signals have not been converted for all pixels, the process returns to step S4. That is, the processes in steps S4 to S10 are repeated until the signals of all the pixels are converted.
[0068]
If it is determined in step S10 that the signals have been processed (converted) for all pixels, the prediction calculation unit 56 outputs the generated component signal to the matrix circuit 36 in step S11. In step S <b> 12, the matrix circuit 36 converts the component signal into an RGB signal and outputs it to the display device 37.
[0069]
In the above, an example in which the class tap is two pixels, that is, when the class tap is vector quantized has been described as being converted into a two-dimensional vector, but a larger number of class taps is set. In some cases, the dimension of the vector to be quantized is only set to a number proportional to the number of class taps. For other points, the composite signal can be converted into a component signal by the same processing as described above.
[0070]
In addition, the class tap setting that determines the feature amount of the pixel of interest has been required to correspond to the phase of the pixel of interest, but it is possible to calculate between pixels of different phases by vector quantization. Therefore, it is possible to set class taps considering all four phases, and it is possible to set a class tap optimal for the pixel of interest.
[0071]
Furthermore, in the conventional class tap setting method, the number of classes of the pixel of interest depends on the number of class taps. For example, when the number of class taps increases, the number of classes also increases. When the number of classes increases in this way, the conventional method reduces the image quality degradation that occurs during conversion, but unnecessarily improves the accuracy and increases the amount of processing, resulting in a burden on the hardware. It was necessary to consider that it would increase. On the other hand, when using representative vectors as described above, the number of representative vectors set in advance is the number of classes itself, so even if the number of classes is determined in advance, an image in which an appropriate representative vector according to the number of classes is input The image quality degradation can be effectively suppressed even with a small number of classes, and the number of classes is not increased unnecessarily, so the processing amount is not increased unnecessarily. The burden on hardware can be reduced, and the processing speed can be improved.
[0072]
Next, referring to FIG. 12, a representative vector setting device 61 that sets a representative vector stored in advance in the representative vector memory 54 and a learning device 62 that generates a prediction coefficient stored in advance in the coefficient memory 57 by learning. explain.
[0073]
The representative vector setting device 61 extracts the setting tap of the composite signal, generates a representative vector group from the pixel values, and outputs and stores the representative vector group in the representative vector memory 54 of the learning device 62 and the class classification adaptive processing circuit 35. The learning device 62 generates a prediction coefficient group based on the composite signal as the student image, the component signal as the teacher image, and the representative vector group input from the representative vector setting device 61, and the class classification adaptive processing circuit 35. Are output to the coefficient memory 57 and stored. Note that the process indicated by the dotted arrow in the figure is the process executed first, and the process indicated by the solid arrow indicates the subsequent process. Therefore, after the representative vector setting process is performed, the prediction coefficient calculation process is performed.
[0074]
Next, the configuration of the representative vector setting device 61 will be described with reference to FIG. The setting tap extraction unit 81 of the representative vector setting device 61 extracts a setting tap (substantially the same as the class tap) from the input composite signal, similarly to the class tap extraction unit 51 of FIG. The data is output to the generation unit 82. Similar to the vector generation unit 52 in FIG. 8, the vector generation unit 82 generates a vector having the pixel value of each pixel extracted as the setting tap, which is input from the setting tap extraction unit 81, as an element, and vector space data Output to the holding unit 83. The vector space data holding unit 83 stores vectors corresponding to all the target pixels of the input composite signal. The representative vector calculation unit 84 obtains a set number of representative vectors from the distribution of vectors stored in the vector space data holding unit 83 (for example, by the LGB algorithm), and stores it in the representative vector memory 54 (FIG. 8). Output and memorize.
[0075]
Next, a representative vector setting process when the representative vector setting device 61 sets a representative vector will be described with reference to the flowchart of FIG.
[0076]
In step S <b> 21, the setting tap extraction unit 81 extracts setting taps from the input composite signal and outputs them to the vector generation unit 82. In step S <b> 22, the vector generation unit 82 generates a vector from the setting tap input from the setting tap extraction unit 81 and outputs the vector to the vector space data holding unit 83. More specifically, the vector generation unit 82 generates a vector having the pixel value of each pixel of the setting tap as an element (vector quantization) and outputs the vector to the vector space data holding unit 83.
[0077]
In step S23, it is determined whether or not a vector has been generated for all the pixels of the input composite signal. If it is determined that no vector has been generated for all the pixels, the process returns to step S21. . That is, the processes in steps S21 to S23 are repeated until a vector is generated for all pixels.
[0078]
If it is determined in step S23 that vectors have been generated for all pixels, the process proceeds to step S24.
[0079]
In step S24, the representative vector calculation unit 84 executes a representative vector determination process. In step S <b> 25, the representative vector calculation unit 84 stores the representative vector obtained by calculation in the representative vector memory 54.
[0080]
Here, the representative vector determination process will be described with reference to the flowchart of FIG. In the following, a case where two representative vectors are set will be described, but the same applies to cases where more than one representative vector is set.
[0081]
In step S41, the representative vector calculation unit 84 initializes a built-in counter n that counts vectors stored in the built-in vector space data holding unit 83 to zero. In the following description, the number of vectors stored in the vector space data holding unit 83 is N.
[0082]
In step S42, the representative vector calculation unit 84 sets the temporary representative vectors rv21 and 22 on the vector space. In step S42, the set temporary representative vectors rv21 and 22 are set to initial values.
[0083]
In step S43, the representative vector calculation unit 84, as shown in FIG. 16, the vector vn (vectors v0 to vN stored in the vector space data holding unit 83) stored in the vector space data holding unit 83, The distances d1 and d2 with the temporary representative vectors rv21 and 22 are obtained.
[0084]
In step S44, the representative vector calculation unit 84 determines whether or not the distance d1 is smaller than the distance d2, for example, when it is determined that the distance d1 is smaller than the distance d2, that is, the vector vn is a temporary representative. If it is determined that the vector is close to the vector rv21, the process proceeds to step S45. In step S45, the representative vector determination unit 84 counts the vector vn as a vector belonging to the set S1 of the representative vectors rv21 as shown in FIG.
[0085]
If the distance d1 is not smaller than the distance d2 in step S44, that is, if the distance d1 is not less than the distance d2 (when it is determined that the vector vn is close to the temporary representative vector rv22), step S46 is performed. In FIG. 17, the representative vector calculation unit 84 counts the vector vn as a vector belonging to the set S2 of the representative vector rv22, as shown in FIG.
[0086]
In step S47, the representative vector calculation unit 84 increments the counter n by 1. In step S48, it is determined whether or not the counter n is equal to or greater than N + 1, that is, whether or not all the vectors stored in the vector space data holding unit 83 are set in any of the sets S1 and S2.
[0087]
For example, if it is determined in step S48 that the counter n is not equal to or greater than N + 1, there will be a vector vn that is not set in any of the sets S1 and S2, and therefore the process returns to step S43. That is, the processes in steps S43 to S48 are repeated until all the vectors are set to one of the sets S1 and S2.
[0088]
If it is determined in step S48 that the counter n is equal to or greater than N + 1, in step S49, the representative vector calculation unit 84, as shown in FIG. 17, the vector rv21 that is the center of gravity position of each distribution of the sets S1 and S2. Find ', 22'.
[0089]
In step S50, the representative vector calculation unit 84 obtains the distance between the vectors rv21 ′ and 22 ′, which are the centroid positions of the sets S1 and S2, and the temporary representative vectors rv21 and 22 and is less than a predetermined value. If it is determined that it is not less than the predetermined value, the process proceeds to step S51. In step S51, the representative vector determining unit 84 replaces the barycentric positions rv21 ′ and 22 ′ with the temporary representative vectors rv21 and 22 and further initializes the vector space data holding unit 83 (deletes stored vector information). The process returns to step S43.
[0090]
If it is determined in step S50 that the distance between the vectors rv21 ′ and 22 ′, which are the gravity center positions of the sets S1 and S2, and the temporary representative vectors rv21 and 22 are equal to or less than a predetermined value, in step S52 The vectors rv21 ′ and 22 ′, which are the centroid positions of the sets S1 and S2, are set as representative vectors, and the processing ends.
[0091]
That is, the distance between the temporary representative vector and each vector is obtained until the difference between the temporary representative vector and the center of gravity of each of the obtained vector set converges to a predetermined value or less. Repeat the setting process. By this process, the representative vector and the representative vector in the state where the variance of the vector distance of each pixel is the smallest are set.
[0092]
In the above processing, the case where there are two class taps and the number of representative vectors is two, but the number of class taps may be m (m> 2). In this case, an m-dimensional vector is set. Further, since the number of representative vectors is the number of classes, an appropriate number of classes may be set and a corresponding number of representative vectors may be set.
[0093]
With the above processing, as shown in FIG. 18, a representative vector corresponding to the distribution of the input composite signal is set. In the example of FIG. 18, there are 11 representative vectors. The class code determination process in the class classification adaptive processing circuit 35 is a process performed by forming a vector formed from the class tap for each pixel of interest and replacing it with the pixel of the nearest representative vector. Since the variance of the distance from the pixel vector is set to the smallest position, even if each pixel is replaced with a representative vector, the difference in the vector space is minimized. It can be carried out. In addition, image quality degradation that occurs when converting a composite signal to a component signal by determining a class code using a representative vector determined by a class tap that considers all four phases of pixels regardless of the phase of the pixel of interest Can be suppressed.
[0094]
Next, a learning device (prediction coefficient computing device) for determining the prediction coefficient stored in the coefficient memory 57 will be described with reference to FIG.
[0095]
The NTSC encoder 101 receives a component signal including a luminance signal Y and color difference signals RY and BY as a teacher image. The NTSC encoder 101 generates an NTSC composite video signal as a student image from the input component signal, and outputs it to the class tap extraction unit 102 and the prediction tap extraction unit 107.
[0096]
The class tap extraction unit 102 extracts pixels (class taps) necessary for class classification from the composite video signal converted into the input digital signal, and outputs the extracted pixels to the vector generation unit 103. The vector generation unit 103 generates a vector from the input class tap. More specifically, the vector generation unit 103 generates a vector using the pixel value of each pixel input as a class tap as a vector element, and outputs the vector to the class code determination unit 104.
[0097]
The class code determination unit 104 searches the representative vector recorded in the representative vector memory 105 for the representative vector closest to the vector input from the vector generation unit 103, and determines the class code corresponding to the searched representative vector. And output to the normal equation generation unit 106. The representative vector memory 105 stores the number of representative vectors of the order corresponding to the input vector by the number corresponding to the class by the above-described processing.
[0098]
The prediction tap extraction unit 107 extracts a prediction tap from the composite video signal input from the NTSC encoder 101 and outputs the prediction tap to the normal equation generation unit 106. The above class tap extraction unit 102, vector generation unit 103, class code determination unit 104, representative vector memory 105, and prediction tap extraction unit 107 are the same as the class tap extraction unit 51 of the class classification adaptive processing circuit 35 in FIG. The generation unit 52, the class code determination unit 53, the representative vector memory 54, and the prediction tap extraction unit 55 have basically the same configuration and function.
[0099]
The normal equation generation unit 106, for all classes input from the class code determination unit 104, for each class, the prediction tap of the student image input from the prediction tap extraction unit 107 and the component signal as the teacher image. A normal equation is generated from the luminance signal Y and output to the coefficient determination unit 108. When the necessary number of normal equations are supplied from the normal equation generation unit 106, the coefficient determination unit 108 solves the normal equations using, for example, the least square method, and calculates the prediction coefficients w1,. Calculate and determine.
[0100]
Here, the normal equation will be described. In the above equation (1), the prediction coefficients w1,..., Wn are undetermined coefficients before learning. Learning is performed by inputting a plurality of teacher images for each class. When the number of types of teacher images is expressed as m, the following equation (2) is set from equation (1).
[0101]
yk = w1 * xk1 + w2 * xk2 ++ ... + wn * xkn (2)
[0102]
Here, k is k = 1, 2,..., M. When m> n, the prediction coefficients w1,..., wn are not uniquely determined. Therefore, the element ek of the error vector e is defined by the following equation (3), and the prediction coefficient is determined so that the error vector e defined by the equation (4) is minimized. That is, for example, the prediction coefficient is uniquely determined by a so-called least square method.
[0103]
ek = yk- {w1 * xk1 + w2 * xk2 ++ ... + wn * xkn} (3)
[0104]
[Expression 1]

[0105]
Each prediction coefficient wi that minimizes e ^ 2 (e squared) in Expression (4) is obtained by partial differentiation of e ^ 2 with the prediction coefficient w i (i = 1, 2,...) (5) The partial differential value is calculated to be 0 for each value of i.
[0106]
[Expression 2]

[0107]
A specific procedure for determining each prediction coefficient w i from equation (5) will be described. When Xji and Yi are defined as in Expression (6) and Expression (7), Expression (5) is transformed into the form of the determinant of Expression (8).
[0108]
[Equation 3]

[0109]
[Expression 4]

[0110]
[Equation 5]

[0111]
This equation (8) is generally called a normal equation. Here, Xji (j, i = 1, 2,... N) and Yi (i = 1, 2,... N) are calculated based on the teacher image and the student image. That is, the normal equation generation unit 106 calculates the values of Xji and Yi and determines the equation (8) consisting of the normal equation. Further, the coefficient determination unit 108 determines each prediction coefficient w i by solving the equation (8).
[0112]
Next, a process in which the learning device in FIG. 19 learns a prediction coefficient will be described with reference to the flowchart in FIG.
[0113]
In step S <b> 71, a component signal as a teacher image is input to the NTSC encoder 101 and the normal equation generation unit 106. In step S <b> 72, the NTSC encoder 101 generates a student image including an NTSC composite signal from the input component signal, and outputs the student image to the class tap extraction unit 102 and the prediction tap extraction unit 107.
[0114]
In step S <b> 73, the class tap extraction unit 102 extracts class taps and outputs them to the vector generation unit 103. This process is the same as the process in step S4 in the flowchart of FIG.
[0115]
In step S74, the vector generation unit 52 reads the pixel values of the class tap extracted by the class tap extraction unit 102, generates a vector composed of these values (vector quantization), and sends the vector to the class code determination unit 104. Output. This process is the same as the process in step S5 in the flowchart of FIG.
[0116]
In step S75, the class code determination unit 104 compares the input vector with a representative vector group stored in advance in the representative vector memory 105, determines the closest representative vector, and predicts the corresponding class (index) code. The result is output to the calculation unit 106.
[0117]
In step S <b> 76, the prediction tap extraction unit 55 extracts the prediction tap of the student image and outputs it to the normal equation generation unit 106.
[0118]
In step S77, the normal equation generation unit 106 uses the above-described equation (5) based on the class code input from the class code determination unit 104, the prediction tap input from the prediction tap extraction unit 107, and the component signal input as a teacher image. The normal equation shown in 8) is generated and output to the coefficient determination unit 108 together with the class code input from the class code determination unit 104.
[0119]
In step S <b> 78, the coefficient determining unit 108 solves the normal equation composed of the above-described equation (8), determines the prediction coefficient, outputs it in association with the class code, and stores it in the coefficient memory 57.
[0120]
In step S79, it is determined whether or not processing has been performed for all pixels. If it is determined that processing has not been performed for all pixels, the processing returns to step S73. That is, the processes in steps S73 to S79 are repeated until the processes for all the pixels are completed. If it is determined in step S79 that all pixels have been processed, the process ends.
[0121]
By the above processing, the class is determined by using a vector obtained by vector quantization of a class tap that is not limited to the target pixel and the surrounding phase, and the class is determined, and the prediction coefficient is calculated. It is possible to obtain a prediction coefficient that can suppress image quality deterioration caused by the conversion process from the composite signal to the component signal. In the above description, the example of setting the prediction coefficient for obtaining the luminance signal Y has been described. However, the color difference signals RY and BY can be obtained in the same manner.
[0122]
Next, referring to FIG. 21 to FIG. 28, SNR (Signal to Noise) at the time of conversion from a composite signal to a component signal by a conventional image conversion apparatus and a television receiver to which the present invention is applied. A comparison result of the ratio between the true component signal corresponding to the signal and the converted component signal will be described.
[0123]
21 and 22, the horizontal direction indicates the number of frames, the vertical direction is SNR (dB), the SNR of the television receiver to which the present invention is applied is a solid line, and the conventional image conversion apparatus The SNR is shown as a dotted line. In the television receiver to which the present invention is applied, as shown in FIG. 23, the target pixel exists on the field # 0, and the class tap has five horizontal pixels centered on the target pixel. A total of 15 taps (number of classes = 32) consisting of 5 pixels in the horizontal direction with the upper and lower pixels as the center, and the prediction tap is the case where the 12 taps shown in FIG. 4 are used, and the conventional image conversion In the apparatus, when the pixel of interest exists on the field # 0, the class tap has the pixel of interest of 5 pixels (number of classes = 31) shown in FIG. 3 and the prediction tap of 12 shown in FIG. The case of tapping is shown.
[0124]
Furthermore, in FIG. 21, an image including many still image areas (704 pixels × 484 pixels, 60 frames, Y + Q), and in FIG. 22, an image including many moving image areas (704 pixels × 484 pixels, 60 frames, Y + Q). ) For each luminance signal Y.
[0125]
In FIG. 21, in the case of a conventional image conversion apparatus, the SNR is around 37.5 dB, whereas in the case of a television receiver to which the present invention is applied, the SNR is around 38 dB. It is shown that the SNR of about 0.5 dB is improved.
[0126]
As for FIG. 22, in the case of the conventional image conversion apparatus and the case of the television receiver to which the present invention is applied, the SNR of the conventional case is more when the present invention is applied to about 1 to 3 dB as a whole. It has been shown to be improved.
[0127]
24 and 25, as in FIGS. 21 and 22, the horizontal direction indicates the number of frames, the vertical direction is SNR, and the SNR of the television receiver to which the present invention is applied is a solid line. In addition, the SNR of the conventional image conversion apparatus is indicated by a dotted line. At this time, in the conventional image conversion apparatus, as shown in FIG. 26, the class tap is 4 in phase with the target pixel on the field # -2 two fields before the class tap shown in FIG. A total of 9 taps (number of classes = 511) including pixels are used, and as shown in FIG. 27, the prediction tap is the 12 taps shown in FIG. 6 pixels consisting of left and right pixels centered on the upper and lower pixels, and the pixel at the target pixel position on the previous field # -2, its left and right pixels, and the upper and lower positions of the target pixel A total of 27 taps (including offset taps), which is a sum of 9 pixels consisting of the pixels to be processed and the left and right pixels, are used.
[0128]
Further, in the television receiver to which the present invention is applied, as shown in FIG. 28, the class tap includes pixels above and below the target pixel position on the field # -1 immediately before the class tap shown in FIG. A total of 6 pixels consisting of the left and right pixels, plus the pixel of interest in the previous field # -2, its left and right pixels, and pixels above and below the pixel of interest, and 9 pixels of the respective left and right pixels 30 taps (number of classes = 512) were used, and the prediction taps used were the same as those in the conventional image conversion apparatus (FIG. 27).
[0129]
24 shows an image (704 pixels × 484 pixels, 60 frames, Y + Q) including many still image regions, and FIG. 25 shows an image including many moving image regions (704 pixels × 484 pixels, 60 frames, Y + Q). These are the results for each luminance signal Y.
[0130]
In FIG. 24, the SNR of the conventional image conversion apparatus is around 43 dB, whereas the SNR of the television receiver to which the present invention is applied is around 43.5 dB, which is higher than that of the conventional image conversion apparatus. The SNR of about 0.5dB has been improved. In addition, in FIG. 25, an improvement in SNR of about 0.5 dB is observed as a whole.
[0131]
Although not shown, it has been confirmed that the color difference signal and other phase signals can be improved in the same manner as in FIG. 21, FIG. 22, FIG. 24, and FIG.
[0132]
As described with reference to FIGS. 21 to 28, a class using a representative vector is used regardless of the number of class taps, an image with many still image regions, or an image with many moving image regions. It is shown that the classification adaptation process effectively improves the SNR at the time of conversion from the composite signal to the component signal, and as a result, the image quality deterioration is suppressed.
[0133]
Further, the above-described vector may be one obtained by normalizing each element. That is, for example, when a vector such as (X100, X101, X102... X200) exists, the maximum value Max (X100, X101, X102... X200) of each of these elements is obtained, and each element is determined. (X100 / Max (X100, X101, X102... X200), X101 / Max (X100, X101, X102... X200), X102 / Max (X100, X101, X102... X200)) ,... X200 / Max (X100, X101, X102... X200)) may be normalized and expressed.
[0134]
When such a normalized vector is used, the index shown in FIG. 9 is expressed as shown in FIG.
[0135]
In the above example, the component video signal (Y, Cr, Cb) is generated from the composite video signal. For example, as shown in FIG. It is also possible to directly generate the primary color RGB signal from the composite video signal output from the above. Also in this case, the configuration of the class classification adaptive processing circuit 121 is the same as that of the class classification adaptive processing circuit 35, but the student image in the learning process for determining the prediction coefficient stored in the coefficient memory 57 is a composite signal. However, the teacher image becomes the primary color RGB signal, and the determined prediction coefficient also corresponds to the primary color RGB signal. However, the representative vector stored in the representative vector memory 54 may be left as it is.
[0136]
Further, the example using the NTSC composite signal has been described above. However, for example, the same configuration can be used when converting a PAL (Phase Alternating Line) composite signal into a component signal or a primary color RGB signal. Can be realized.
[0137]
According to the above, since the class code is determined based on the representative vector, it is possible to classify the pixel of interest appropriately, taking into account the relative relations for all phases of the composite signal. It is possible to suppress image quality degradation that occurs during conversion to a signal.
[0138]
The series of processes described above can be executed by hardware, but can also be executed by software. When a series of processes is executed by software, a program constituting the software may execute various functions by installing a computer incorporated in dedicated hardware or various programs. For example, it is installed from a recording medium in a general-purpose personal computer or the like.
[0139]
FIG. 31 shows a configuration of an embodiment of a personal computer when the television receiver, the representative vector setting device, or the learning device is realized by software. The CPU 201 of the personal computer controls the entire operation of the personal computer. Further, when a command is input from the input unit 206 such as a keyboard or a mouse from the user via the bus 204 and the input / output interface 205, the CPU 201 is stored in a ROM (Read Only Memory) 202 correspondingly. Run the program. Alternatively, the CPU 201 reads a program read from the magnetic disk 211, the optical disk 212, the magneto-optical disk 213, or the semiconductor memory 214 connected to the drive 210 and installed in the storage unit 208 into a RAM (Random Access Memory) 203. To load and execute. Thereby, the functions of the above-described image processing apparatus are realized by software. Further, the CPU 201 controls the communication unit 209 to communicate with the outside and exchange data.
[0140]
As shown in FIG. 31, the recording medium on which the program is recorded is distributed to provide the program to the user separately from the computer, and the magnetic disk 211 (including the flexible disk) on which the program is recorded, By a package medium comprising an optical disk 212 (including compact disc-read only memory (CD-ROM), DVD (digital versatile disc)), a magneto-optical disk 213 (including MD (mini-disc)), or a semiconductor memory 214 In addition to being configured, it is configured by a ROM 202 on which a program is recorded, a hard disk included in the storage unit 208, and the like provided to the user in a state of being incorporated in a computer in advance.
[0141]
In this specification, the step of describing the program recorded on the recording medium is not limited to the processing performed in time series in the order described, but of course, it is not necessarily performed in time series. Or the process performed separately is included.
[0142]
【The invention's effect】
  According to the image information conversion apparatus and method and the first program of the present invention, the pixel of interestTo classifyExtracting a plurality of first pixel signals from the input composite signal;Vector quantization of the pattern of multiple pixel signals for classifying each pixel of the input composite signal into a class, and using that in the vector spacePre-statistical learningPer classStores the specified representative pattern, searches for the closest representative pattern to the extracted first pixel signal patterns, and searches for the representative patternThe class where,Classify into pixel class of interestFrom the input composite signalFor converting the pixel of interestA plurality of second pixel signals are extracted, and the second pixel signals and classified classesSet for eachWith coefficientSum of productsThe pixel signal of the pixel of interest is converted by calculation.
[0144]
  Coefficient calculation apparatus and method of the present invention, and2According to the program, the input composite signal is generated from the input image signal, and the target pixelTo classifyExtracting pixel signals of a plurality of first pixels from an input composite signal;Vector quantization of the pattern of multiple pixel signals for classifying each pixel of the input composite signal into a class, and using that in the vector spacePre-statistical learningPer classStores the specified representative pattern, searches for the closest representative pattern to the extracted first pixel signal patterns, and searches for the representative patternSaid class is defined,Classify into pixel class of interestAnd extracting a plurality of second pixel signals from the input composite signal and inputting the second pixel signals and the image signalFrom the normal equation using andCategorized classEveryThe coefficient was calculated.
[0145]
  Representative pattern setting apparatus and method of the present invention, and3According to the program, each pixel of the input composite signalTo classifyExtract multiple pixel signals, vector quantize multiple pixel signals, calculate representative vector in vector space according to the number of classes from multiple vector quantized pixel signals, and calculate representative vector as representative pattern Was set in association with the class.
[0146]
As a result, in both cases, it is possible to classify the pixel of interest appropriately, taking into account the relative relationships of all phases of the composite signal, and to suppress image quality degradation that occurs during conversion from the composite signal to the component signal. It becomes possible.
[Brief description of the drawings]
FIG. 1 is a block diagram illustrating a configuration example of a conventional image conversion apparatus.
FIG. 2 is a diagram for explaining an NTSC composite signal.
FIG. 3 is a diagram illustrating class taps.
FIG. 4 is a diagram for explaining a prediction tap.
FIG. 5 is a block diagram illustrating a configuration example of a conventional learning device.
FIG. 6 is a block diagram illustrating a configuration of an embodiment of a television receiver to which the present invention has been applied.
7 is a diagram for explaining a class classification adaptive processing circuit in FIG. 6; FIG.
8 is a block diagram illustrating a configuration of a class classification adaptive processing circuit in FIG. 6. FIG.
FIG. 9 is a diagram for explaining representative vectors stored in the representative vector memory of FIG. 8;
10 is a flowchart illustrating display processing of the television receiver in FIG. 6. FIG.
FIG. 11 is a diagram illustrating processing for selecting a representative vector.
FIG. 12 is a diagram illustrating a representative vector setting device and a learning device.
FIG. 13 is a block diagram illustrating a configuration of a representative vector setting device.
14 is a flowchart for explaining representative vector setting processing of the representative vector setting device of FIG. 13;
15 is a flowchart for explaining representative vector determination processing by the representative vector calculation unit in FIG. 13;
FIG. 16 is a diagram illustrating a method for calculating a representative vector.
FIG. 17 is a diagram illustrating a method for calculating a representative vector.
FIG. 18 is a diagram illustrating a method for calculating a representative vector.
FIG. 19 is a block diagram illustrating a learning device.
FIG. 20 is a flowchart illustrating a learning process of the learning device.
FIG. 21 is a diagram showing a comparison result of SNR between a conventional image conversion apparatus and a television receiver to which the present invention is applied.
FIG. 22 is a diagram showing a comparison result of SNR between a conventional image conversion apparatus and a television receiver to which the present invention is applied.
FIG. 23 is a diagram illustrating a class tap of a television receiver to which the present invention is applied as in FIGS. 21 and 22.
FIG. 24 is a diagram showing a comparison result of SNR between a conventional image conversion apparatus and a television receiver to which the present invention is applied.
FIG. 25 is a diagram showing a comparison result of SNR between a conventional image conversion apparatus and a television receiver to which the present invention is applied.
FIG. 26 is a diagram showing a class tap of the conventional image conversion apparatus in FIGS. 24 and 25.
FIG. 27 is a diagram illustrating prediction taps in FIGS. 24 and 25;
FIG. 28 is a diagram showing a class tap of a television receiver to which the present invention is applied as in FIGS. 24 and 25.
FIG. 29 is a diagram for explaining an example when the representative vector in FIG. 9 is normalized;
FIG. 30 is a block diagram illustrating another configuration of a television receiver to which the present invention has been applied.
FIG. 31 is a diagram illustrating a medium.
[Explanation of symbols]
35 class classification adaptive processing circuit, 51 class tap extraction unit, 52 vector generation unit, 53 class code determination unit, 54 representative vector memory, 55 prediction tap extraction unit, 56 prediction calculation unit, 57 coefficient memory, 61 representative vector setting device, 62 learning device, 81 setting tap extracting unit, 82 vector generating unit, 83 vector space data holding unit, 84 representative vector calculating unit, 101 NTSC encoder, 102 class tap extracting unit, 103 vector generating unit, 104 class code determining unit, 105 Representative vector memory, 106 normal equation generation unit, 107 prediction tap extraction unit, 108 coefficient determination unit, 121 class classification adaptive processing circuit

Claims (19)

First extraction means for extracting a plurality of first pixel signals for classifying the pixel of interest into a class from an input composite signal;
A vector pattern of a plurality of pixel signals for classifying each pixel of the input composite signal into the class is vector-quantized, and a representative pattern previously determined for each class is obtained by statistical learning on a vector space using the pattern. Representative pattern storage means for storing;
Representative pattern search means for searching for the representative pattern closest to the pattern of the plurality of first pixel signals extracted by the first extraction means;
Class classification means for classifying the class in which the representative pattern searched by the representative pattern search means is defined into a class of the pixel of interest ;
Second extraction means for extracting a plurality of second pixel signals for converting the pixel of interest from the input composite signal;
Pixel-of- interest signal conversion means for converting the pixel signal of the pixel of interest by performing a product-sum operation on the second pixel signal and the coefficient set for each class classified by the class classification means. An image information conversion device characterized by the above. Vector quantization means for vector-quantizing the plurality of first pixel signals;
The representative pattern storage means, among the plurality of first pixel signal for classifying the pixel of interest of the composite signal to a class, a representative pattern consisting of a representative of the plurality of first pixel signals as a representative vector The image information conversion device according to claim 1, wherein the image information conversion device is stored. The representative pattern search means includes a vector quantization of the plurality of first pixel signals extracted by the first extraction means, and the representative pattern as the representative vector closest to the vector space, The image information conversion apparatus according to claim 2, wherein a search is performed as the representative pattern closest to the pattern of the plurality of first pixel signals extracted by the first extraction unit. The pixel-of- interest signal conversion unit converts the pixel signal of the pixel of interest into a component signal by performing a product-sum operation on the second pixel signal and the coefficient set in the class classified by the class classification unit. The image information conversion apparatus according to claim 1, wherein: The pixel-of- interest signal converter converts the pixel signal of the pixel of interest into a luminance signal, or a product-sum operation of the second pixel signal and the coefficient set in the class classified by the class classification unit, or The image information conversion apparatus according to claim 1, wherein the image information conversion apparatus converts to a chroma signal.   Display means for displaying the pixel signal of the pixel of interest converted into the component signal is further provided.
  The image information conversion apparatus according to claim 1. A first extraction step of extracting a plurality of first pixel signals for classifying the pixel of interest into a class from an input composite signal;
A vector pattern of a plurality of pixel signal patterns for classifying each pixel of the input composite signal into the class is vector-quantized, and a representative pattern predetermined for each class is obtained by statistical learning on a vector space using the pattern. A representative pattern storing step for storing;
A representative pattern search step of searching for the representative pattern closest to the turn of the plurality of first pixel signals extracted in the processing of the first extraction step;
A class classification step for classifying the class in which the representative pattern searched in the processing of the representative pattern search step is defined into a class of the target pixel ;
A second extraction step of extracting a plurality of second pixel signals for converting the target pixel from the composite signal;
A pixel-of- interest signal conversion step of converting a pixel signal of the pixel of interest by a product-sum operation of the second pixel signal and a coefficient set for each of the classes classified in the processing of the class classification step; An image information conversion method comprising: A first extraction control step for controlling extraction of a plurality of first pixel signals for classifying the pixel of interest into a class from an input composite signal;
A vector pattern of a plurality of pixel signals for classifying each pixel of the input composite signal into the class is vector-quantized, and a representative pattern predetermined for each class is obtained by statistical learning on a vector space using the pattern. Representative pattern memory control step for controlling memory;
A representative pattern search control step for controlling the search of the representative pattern closest to the pattern of the plurality of first pixel signals whose extraction is controlled in the processing of the first extraction control step;
A class classification control step for controlling classification in which the class in which the representative pattern searched in the processing of the representative pattern search control step is defined is the class of the pixel of interest ;
A second extraction control step for controlling extraction of a plurality of second pixel signals for converting the target pixel from the input composite signal;
The pixel of interest that controls the conversion of the pixel signal of the pixel of interest by the product-sum operation of the second pixel signal and the coefficient set for each class whose classification is controlled in the processing of the class classification control step A recording medium on which a computer-readable program is recorded, comprising: a signal conversion control step. A first extraction control step for controlling extraction of a plurality of first pixel signals for classifying the pixel of interest into a class from an input composite signal;
A representative pattern determined for each class is obtained by vector quantization of a plurality of pixel signal patterns for classifying each pixel of the input composite signal into the class, and statistical learning on a vector space using the pattern. Representative pattern storage control step for controlling
A representative pattern search control step for controlling the search of the representative pattern closest to the pattern of the plurality of first pixel signals whose extraction is controlled in the processing of the first extraction control step;
A class classification control step for controlling classification in which the class in which the representative pattern searched in the processing of the representative pattern search control step is defined is the class of the pixel of interest ;
A second extraction control step for controlling extraction of a plurality of second pixel signals for converting the target pixel from the input composite signal;
The pixel of interest for controlling the conversion of the pixel signal of the pixel of interest by the product-sum operation of the second pixel signal and the coefficient set for each class whose classification is controlled in the processing of the class classification control step A program that causes a computer to execute a signal conversion control step. Composite signal generating means for generating an input composite signal from the input image signal;
First extraction means for extracting pixel signals of a plurality of first pixels for classifying the pixel of interest into a class from the input composite signal;
A vector pattern of a plurality of pixel signals for classifying each pixel of the input composite signal into the class is vector-quantized, and a representative pattern previously determined for each class is obtained by statistical learning on a vector space using the pattern. Representative pattern storage means for storing;
Representative pattern search means for searching for the representative pattern closest to the pattern of the plurality of first pixel signals extracted by the first extraction means;
Class classification means for classifying the class in which the representative pattern searched by the representative pattern search means is defined into a class of the pixel of interest ;
Second extraction means for extracting a plurality of second pixel signals for converting the pixel of interest from the input composite signal;
A coefficient calculation unit comprising: a calculation unit configured to calculate a coefficient for each of the classes classified by the class classification unit based on a normal equation using the second pixel signal and the input image signal. . A storage device that stores the coefficient calculated by the coefficient calculation device according to claim 10 . A composite signal generation step for generating an input composite signal from the input image signal;
A first extraction step of extracting pixel signals of a plurality of first pixels for classifying the pixel of interest into a class from the input composite signal;
A vector pattern of a plurality of pixel signal patterns for classifying each pixel of the input composite signal into the class is vector-quantized, and a representative pattern predetermined for each class is obtained by statistical learning on a vector space using the pattern. A representative pattern storing step for storing;
A representative pattern search step of searching for the representative pattern closest to the pattern of the plurality of first pixel signals extracted in the processing of the first extraction step;
A class classification step of classifying the class in which the representative pixel signal searched in the processing of the representative pattern search step is defined into a class of the pixel of interest ;
A second extraction step of extracting a plurality of second pixel signals for converting the pixel of interest from the input composite signal;
A coefficient step of calculating a coefficient for each of the classes classified in the class classification step from a normal equation using the second pixel signal and the input image signal. Calculation method. A composite signal generation control step for controlling generation of an input composite signal from an input image signal;
A first extraction control step for controlling extraction of a pixel signal from the input composite signal of a plurality of first pixels for classifying the pixel of interest into a class ;
A vector pattern of a plurality of pixel signals for classifying each pixel of the input composite signal into the class is vector-quantized, and a representative pattern predetermined for each class is obtained by statistical learning on a vector space using the pattern. Representative pattern memory control step for controlling memory;
A representative pattern search control step for controlling the search of the representative pattern closest to the pattern of the plurality of first pixel signals whose extraction is controlled in the processing of the first extraction control step;
A class classification control step for controlling classification in which the class in which the representative pattern searched in the processing of the representative pattern search control step is defined is the class of the pixel of interest ;
A second extraction control step for controlling extraction of a plurality of second pixel signals for converting the target pixel from the input composite signal;
A calculation control step for controlling the calculation of the coefficient for each class whose classification is controlled by the processing of the class classification control step based on a normal equation using the second pixel signal and the input image signal. A recording medium on which a computer-readable program is recorded. A composite signal generation control step for controlling the generation of an input composite signal from the input image signal;
A first extraction control step for controlling extraction of pixel signals of a plurality of first pixels for classifying a pixel of interest into a class from the input composite signal;
A vector pattern of a plurality of pixel signals for classifying each pixel of the input composite signal into the class is vector-quantized, and a representative pattern predetermined for each class is obtained by statistical learning on a vector space using the pattern. Representative pattern memory control step for controlling memory;
A representative pattern search control step for controlling the search of the representative pattern closest to the pattern of the plurality of first pixel signals whose extraction is controlled in the processing of the first extraction control step;
A class classification control step for controlling classification in which the class in which the representative pattern searched in the processing of the representative pattern search control step is defined is the class of the pixel of interest ;
A second extraction control step for controlling extraction of a plurality of second pixel signals from the input composite signal;
A calculation control step for controlling calculation of a predetermined coefficient for each of the classes whose classification is controlled by the processing of the class classification control step based on a normal equation using the second pixel signal and the input image signal; A program that causes a computer to execute. A representative for setting a representative pattern composed of a plurality of representative pixel signals out of the plurality of pixel signals extracted from the input composite signal, which is used when the input composite signal is converted by the class classification adaptive processing. In the pattern setting device,
Extracting means for extracting a plurality of pixel signals for classifying each pixel of the input composite signal into the class ;
Vector quantization means for vector-quantizing the plurality of pixel signals;
Calculation means for calculating a representative vector in a vector space according to the number of classes from the plurality of pixel signals vector quantized by the vector quantization means;
A representative pattern setting device comprising: a setting unit configured to set the representative vector calculated by the calculation unit in association with the class as the representative pattern. Said computing means according to claim 15, wherein the plurality of pixel signals vector quantization by the vector quantization means LBG algorithm, and calculating in accordance with the representative vectors in a vector space with the number of the class The representative pattern setting device described in 1. A representative pattern consisting of pixel signals of a plurality of representative pixels among the pixel signals of a plurality of pixels extracted from the input composite signal, used when converting the input composite signal by class classification adaptive processing, is set. In the representative pattern setting method of the representative pattern setting device,
An extraction step of extracting a plurality of pixel signals for classifying each pixel of the input composite signal into the class ;
A vector quantization step for vector quantization of the plurality of pixel signals;
A calculation step of calculating a representative vector on a vector space according to the number of classes from the plurality of pixel signals vector-quantized by the processing of the vector quantization step;
A representative pattern setting method comprising: a setting step of setting the representative vector calculated in the calculation step in association with the class as the representative pattern. A representative for setting a representative pattern composed of a plurality of representative pixel signals among pixel signals of a plurality of pixels extracted from the input composite signal, which is used when the input composite signal is converted by class classification adaptive processing. A program for controlling a pattern setting device,
An extraction control step for controlling extraction of a plurality of pixel signals for classifying each pixel of the input composite signal into the class ;
A vector quantization control step for controlling vector quantization of the plurality of pixel signals;
A calculation control step for controlling calculation in accordance with the number of classes of representative vectors in a vector space from the plurality of pixel signals whose vector quantization is controlled by the processing of the vector quantization control step;
A setting control step for controlling a setting of the representative vector calculated in the processing of the calculation control step in association with the class as the representative pattern. Recording medium. A representative for setting a representative pattern composed of a plurality of representative pixel signals among pixel signals of a plurality of pixels extracted from the input composite signal, which is used when the input composite signal is converted by class classification adaptive processing. An extraction control step of controlling extraction of a plurality of pixel signals for classifying each pixel of the input composite signal into the class to a computer that controls a pattern setting device;
A vector quantization control step for controlling vector quantization of the plurality of pixel signals;
A calculation control step for controlling calculation in accordance with the number of classes of representative vectors in a vector space from the plurality of pixel signals whose vector quantization is controlled by the processing of the vector quantization control step;
A setting control step for executing a setting control step for controlling a setting of the representative vector calculated in the calculation control step associated with the class as the representative pattern.

Images