JP3786300B2 - Motion vector detection apparatus and motion vector detection method - Google Patents

Description

で表すことができる。
【００２３】
一方、ブロツク差分値演算部１Ｂの上ブロツクベクトル値メモリ６は、格納しておいた上ブロツク動きベクトル値（ｘ ｍ₀ −１、ｙ ｎ₀ ）をベクトル差分重み演算回路１１に送出する。同様に、左ブロツクベクトル値メモリ７は格納しておいた左ブロツク動きベクトル値（ｘ ｍ₀ 、ｙ ｎ₀ −１）をベクトル差分重み演算回路１２に送出する。
【００２４】
ベクトル差分重み演算回路１１は、入力された上ブロツク動きベクトル値（ｘ ｍ₀ −１、ｙ ｎ₀ ）と、ベクトルコントロール回路９から入力される現在演算の行われている評価ブロツク（ｍ_-1、ｎ_-1）から基準ブロツク（ｍ₀ 、ｎ₀ ）までの基準ブロツク動きベクトル値（ｘ、ｙ）との差分（距離）に比例した値を、重み値Ｗ_up（ｘ、ｙ）として、次式
【数２】

で算出するようになされている。
【００２５】
同様に、ベクトル差分重み演算回路１２は、入力された左ブロツク動きベクトル値（ｘ ｍ₀ 、ｙ ｎ₀ −１）と、ベクトルコントロール回路９から入力された基準ブロツク動きベクトル値（ｘ、ｙ）との差分（距離）に比例した値を、重み値Ｗ_left（ｘ、ｙ）として、次式
【数３】

で算出するようになされている。
【００２６】
ここで、（２）式及び（３）式において用いられている係数ｋは、現フレーム用メモリ２から読み出された画像データＤ２の分散値（ばらつき）に応じて変化する値であり、係数演算回路１６によつて基準ブロツクの位置が変わる度に新たに生成され、乗算器１７を介して重み値Ｗ_up（ｘ、ｙ）及びＷ_left（ｘ、ｙ）に乗算されるようになされている。
【００２７】
つまり、係数演算回路１６は画像データＤ２のばらつきが大きい場合、各評価ブロツク（ｍ_-1、ｎ_-1）との差分の絶対値の総和Ｓ_um（ｘ、ｙ）も当然ばらつきが大きくなるので、係数ｋを大きな値とすることにより重み値Ｗ_up（ｘ、ｙ）及びＷ_left（ｘ、ｙ）を大きくして大きく重み付けするようになされている。
また、係数演算回路１６は画像データＤ２のばらつきが小さい場合、各評価ブロツク（ｍ_-1、ｎ_-1）との差分の絶対値の総和Ｓ_um（ｘ、ｙ）も当然ばらつきが小さくなるので、係数ｋを小さな値とすることにより重み値Ｗ_up（ｘ、ｙ）及びＷ_left（ｘ、ｙ）を小さくして小さく重み付けするようになされている。
【００２８】
このように、係数演算回路１６は画像データＤ２のばらつきに応じて係数ｋの値を変動させることにより、評価ブロツク（ｍ_-1、ｎ_-1）のそれぞれの位置における総和Ｓ_um（ｘ、ｙ）の値に重み付けしたときに係数ｋによつて大きく変動させない（打ち消さない）ようにしている。従つて、重み値Ｗ_up（ｘ、ｙ）及びＷ_left（ｘ、ｙ）は総和Ｓ_um（ｘ、ｙ）の値と比べればごく微小な値となる。この場合、重み値Ｗ_up（ｘ、ｙ）及びＷ_left（ｘ、ｙ）は、基準ブロツク動きベクトル値（ｘ、ｙ）が上ブロツク動きベクトル値（ｘ ｍ₀ −１、ｙ ｎ₀ ）及び左ブロツク動きベクトル値（ｘ ｍ₀ 、ｙ ｎ₀ −１）に近ければ近いほど距離が短くなるので差分は小さくなり、当然重み値も小さくなる。
【００２９】
このようにして、算出された重み値Ｗ_up（ｘ、ｙ）及びＷ_left（ｘ、ｙ）は加算器１３によつて加えられ、さらに差分演算回路８から出力される差分の絶対値の総和Ｓ_um（ｘ、ｙ）に加算器１４を介して加えられてブロツク差分値Ｄ_iff （ｘ、ｙ）として、次式
【数４】

として表され、検出手段としてのベクトル値検出回路１５に送出される。
【００３０】
このようにして、サーチエリア内の全ての評価ブロツク（ｍ_-1、ｎ_-1）に対して算出したブロツク差分値Ｄ_iff （ｘ、ｙ）がベクトル値検出回路１５に入力される。ベクトル値検出回路１５は、全てのブロツク差分値Ｄ_iff （ｘ、ｙ）を比較し、当該ブロツク差分値Ｄ_iff （ｘ、ｙ）が最も小さくなるときの基準ブロツク動きベクトル値（ｘ、ｙ）を出力するようになされている。
また、ベクトル値検出回路１５は検出された基準ブロツク動きベクトル値（ｘ、ｙ）を上ブロツクベクトル値メモリ６及び左ブロツクベクトル値メモリ７にも送出し、右及び下ブロツクが次に基準ブロツク（ｍ₀ 、ｎ₀ ）となつたときの動きベクトル値として用いることができるようにしている。
【００３１】
以上の構成において、動きベクトル検出装置１においては図６（Ａ）及び（Ｂ）に示す基準ブロツク（ｍ₀ 、ｎ₀ ）と評価ブロツク（ｍ_-1、ｎ_-1）とに基づいて画像が右側から左側に移動する場合について説明する。
動きベクトル検出装置１では、差分演算部１Ａによつて現フレームＦ₀ の基準ブロツク（ｍ₀ 、ｎ₀ ）の画像データＤ２と前フレームＦ_-1のサーチエリア内の評価ブロツク（ｍ_-1、ｎ_-1）の画像データＤ３との各画素値の差分の絶対値の総和Ｓ_um（ｘ、ｙ）を（１）式によつて算出する。
【００３２】
ここで、サーチエリア内の全ての評価ブロツク（ｍ_-1、ｎ_-1）についての総和Ｓ_um（ｘ、ｙ）を算出し、この算出結果が最小となる評価ブロツク（ｍ_-1、ｎ_-1）の位置を求めただけでは、レンガの連続した同一パターンのブロツク画像が複数存在しているために総和Ｓ_um（ｘ、ｙ）が最小となる評価ブロツク（ｍ_-1、ｎ_-1）の位置を複数検出（図７）してしまつて正確な動きベクトルは検出できない。
【００３３】
そこで、動きベクトル検出装置１はブロツク差分値演算部１Ｂによつて予め記憶させておいた上ブロツク動きベクトル値（ｘ ｍ₀ −１、ｙ ｎ₀ ）及び左ブロツク動きベクトル値（ｘ ｍ₀ 、ｙ ｎ₀ −１）と（図３（Ａ））、基準ブロツク動きベクトル値（ｘ、ｙ）と（図３（Ｂ））の差分の絶対値に係数ｋを乗算した重み値Ｗ_up（ｘ、ｙ）、Ｗ_left（ｘ、ｙ）を（２）式及び（３）式によつて算出し、当該重み値Ｗ_up（ｘ、ｙ）、Ｗ_left（ｘ、ｙ）を（４）式によつて各画素値の差分の絶対値の総和Ｓ_um（ｘ、ｙ）に加算してブロツク差分値Ｄ_iff （ｘ、ｙ）を求める。
【００３４】
実際上、動きベクトル検出装置１では上ブロツク動きベクトル値（ｘ ｍ₀ −１、ｙ ｎ₀ ）、及び左ブロツク動きベクトル値（ｘ ｍ₀ 、ｙ ｎ₀ −１）が固定値であり、サーチエリア内の全ての基準ブロツク動きベクトル値（ｘ、ｙ）が評価ブロツク（ｍ_-1、ｎ_-1）の位置によつて変動する変動値であるために、評価ブロツク（ｍ_-1、ｎ_-1）がサーチエリア内の本来の動きベクトルを表す位置にあつた場合にだけ重み値Ｗ_up（ｘ、ｙ）、Ｗ_left（ｘ、ｙ）が最小になり、かくしてブロツク差分値Ｄ_iff （ｘ、ｙ）は最小になる。
【００３５】
従つて、図４に示すように動きベクトル検出装置１ではサーチエリア内の各評価ブロツク（ｍ_-1、ｎ_-1）のそれぞれの位置におけるブロツク差分値Ｄ_iff （ｘ、ｙ）は、重み値Ｗ_up（ｘ、ｙ）、Ｗ_left（ｘ、ｙ）の大きさに応じた傾斜を持つグラフとして表示する。これにより、動きベクトル検出装置１はブロツク差分値Ｄ_iff （ｘ、ｙ）が最小となる位置の基準ブロツク動きベクトル値（ｘ、ｙ）をベクトル値検出回路１５によつて検出することができ、かくしてレンガのような連続したパターンのブロツク画像であつても動きベクトルを誤りなく検出することができる。
【００３６】
また、動きベクトル検出装置１では重み値Ｗ_up（ｘ、ｙ）、Ｗ_left（ｘ、ｙ）を算出する際に係数ｋの値を基準ブロツク（ｍ₀ 、ｎ₀ ）の画素のばらつきに応じて変動させることにより、総和Ｓ_um（ｘ、ｙ）を係数ｋによつて打ち消すことなく重み付けしてブロツク差分値Ｄ_iff （ｘ、ｙ）を表すグラフの傾斜の角度を変えることができ、かくしてブロツク差分値Ｄ_iff （ｘ、ｙ）が最小となる基準ブロツク動きベクトル値（ｘ、ｙ）の検出を容易にして動きベクトルの検出ミスを低減させることができる。
【００３７】
このように、動きベクトル検出装置１は差分演算部１Ａによつて基準ブロツク（ｍ₀ 、ｎ₀ ）と評価ブロツク（ｍ_-1、ｎ_-1）のそれぞれの位置における画素値の差分の絶対値の総和Ｓ_um（ｘ、ｙ）を算出し、ブロツク差分値演算部１Ｂによつて重み値Ｗ_up（ｘ、ｙ）、Ｗ_left（ｘ、ｙ）を算出し、当該重み値Ｗ_up（ｘ、ｙ）、Ｗ_left（ｘ、ｙ）を評価ブロツク（ｍ_-1、ｎ_-1）のそれぞれの位置における総和Ｓ_um（ｘ、ｙ）に加算してそれぞれのブロツク差分値Ｄ_iff （ｘ、ｙ）を算出し、検出手段であるベクトル値検出回路１５によつてブロツク差分値Ｄ_iff （ｘ、ｙ）が最小となる基準ブロツク動きベクトル値（ｘ、ｙ）を検出することができる。
【００３８】
以上の構成によれば、動きベクトル検出装置１は、基準ブロツク（ｍ₀ 、ｎ₀ ）の画像データＤ２と評価ブロツク（ｍ_-1、ｎ_-1）の画像データＤ３との各画素値の差分の絶対値の総和Ｓ_um（ｘ、ｙ）に、上ブロツク動きベクトル値（ｘ ｍ₀ −１、ｙ ｎ₀ ）及び左ブロツク動きベクトル値（ｘ ｍ₀ 、ｙ ｎ₀ −１）と、基準ブロツク動きベクトル値（ｘ、ｙ）との差分の絶対値に係数ｋをかけた重み値Ｗ_up（ｘ、ｙ）、Ｗ_left（ｘ、ｙ）を加算してブロツク差分値Ｄ_iff （ｘ、ｙ）を算出するようにしたことにより、ブロツク差分値Ｄ_iff （ｘ、ｙ）が最小となる基準ブロツク動きベクトル値（ｘ、ｙ）の検出を容易にし、かくしてブロツクごとの画像パターンが似通つている場合でも動きベクトルの検出ミスを低減させて高品質な画像データを伝送することができる。
【００３９】
また、動きベクトル検出装置１は基準ブロツク（ｍ₀ 、ｎ₀ ）の画像データＤ２の分散値に比例した係数ｋを上ブロツク動きベクトル値（ｘ ｍ₀ −１、ｙ ｎ₀ ）及び左ブロツク動きベクトル値（ｘ ｍ₀ 、ｙ ｎ₀ −１）と、基準ブロツク動きベクトル値（ｘ、ｙ）との差分の絶対値に乗算することにより、画素値の差分の総和Ｓ_um（ｘ、ｙ）に応じた重み値Ｗ_up（ｘ、ｙ）、Ｗ_left（ｘ、ｙ）を算出することができ、かくしてブロツク差分値Ｄ_iff （ｘ、ｙ）が最小となる場合の基準ブロツク（ｍ₀ 、ｎ₀ ）の動きベクトル検出を容易にして検出ミスを低減させることができる。
【００４０】
さらに、動きベクトル検出装置１は基準ブロツク（ｍ₀ 、ｎ₀ ）の周囲に位置する周囲ブロツクのうち、基準ブロツク（ｍ₀ 、ｎ₀ ）の上部に隣接する上ブロツク（ｍ₀ −１、ｎ₀ ）と、左部に隣接する左ブロツク（ｍ₀ 、ｎ₀ −１）とに基づいて重み値Ｗ_up（ｘ、ｙ）、Ｗ_left（ｘ、ｙ）を算出するようにしたことにより、演算を容易にして演算時間を短縮させることができ、かくしてリアルタイムに動きベクトルを検出することができる。
【００４１】
なお上述の実施例においては、動きベクトル検出装置１において、現フレームＦ₀ の基準ブロツク（ｍ₀ 、ｎ₀ ）に対して上部及び左部に隣接する上ブロツク（ｍ₀ −１、ｎ₀ ）及び左ブロツク（ｍ₀ 、ｎ₀ −１）の各動きベクトル値（ｘ ｍ₀ −１、ｙ ｎ₀ ）及び（ｘ ｍ₀ 、ｙ ｎ₀ −１）と、基準ブロツク動きベクトル値（ｘ、ｙ）との差分の絶対値に基づいて重み値Ｗ_up（ｘ、ｙ）、Ｗ_left（ｘ、ｙ）を算出するようにした場合について述べたが、本発明はこれに限らず、図５に示す動きベクトル検出装置３０のように前フレームの動きベクトル値を格納しておく前フレームベクトル値メモリ３１、及びベクトル差分重み演算回路３２を動きベクトル検出装置１に対して付け加えて、前フレームの動きベクトル値と、基準ブロツク動きベクトル値（ｘ、ｙ）との差分の絶対値に基づいて重み値Ｗ_for （ｘ、ｙ）を算出して当該重み値Ｗ_for （ｘ、ｙ）を、重み値Ｗ_up（ｘ、ｙ）、Ｗ_left（ｘ、ｙ）に加えてブロツク差分値Ｄ_iff （ｘ、ｙ）を求めるようにしても良い。この場合にも上述の実施例と同様の効果を得ることができる。
【００４２】
また上述の実施例においては、サーチエリアの面積範囲を基準ブロツク（ｍ₀ 、ｎ₀ ）の面積の約４倍程度にするようにした場合について述べたが、本発明はこれに限らず、画像のパターンや処理時間に応じてサーチエリアの面積範囲を自由に設定するようにしても良い。
【００４３】
さらに上述の実施例においては、ブロツク差分値Ｄ_iff （ｘ、ｙ）を求めるために基準ブロツク（ｍ₀ 、ｎ₀ ）の周囲の上部及び左部に隣接する上ブロツク（ｍ₀ −１、ｎ₀ ）及び左ブロツク（ｍ₀ 、ｎ₀ −１）の各動きベクトル値（ｘ ｍ₀ −１、ｙ ｎ₀ ）及び（ｘ ｍ₀ 、ｙ ｎ₀ −１）を用いるようにした場合について述べたが、本発明はこれに限らず、要は基準ブロツク（ｍ₀ 、ｎ₀ ）に対して既に画像データの処理がなされて動きベクトル値の算出されたブロツクであれば上ブロツクあるいは左ブロツクのどちらか一方の動きベクトル値だけでブロツク差分値Ｄ_iff （ｘ、ｙ）を求めたり、また左斜め上のブロツクや右斜め上のブロツクの動きベクトル値を用いるようにしても良い。
【００４４】
さらに上述の実施例においては、本発明をＨＤＴＶ用のＶＴＲに適用するようにした場合について述べたが、本発明はこれに限らず、デイジタル画像通信の分野において動画像を高能率符号化して伝送する種々の装置に適用するようにしても良い。
【００４５】
【発明の効果】
上述のように本発明によれば、画素値の差分の総和が最小になる評価ブロツクが複数存在するときにも、基準ブロツクの画像データのばらつき傾向を反映させた重み付け量を基に最終的な動きベクトルを検出することができるので、同一絵柄の連続したパターンのブロツク画像であつても誤りなく動きベクトルを検出することができ、かくして同様の画像パターンが連続する画像及び輝度が一定の画像において動きベクトルの検出ミスを低減させ得る動きベクトル検出装置及び動きベクトル検出方法を実現することができる。
【図面の簡単な説明】
【図１】本発明の実施例によるブロツクマツチングを行う場合の説明に供する略線図である。
【図２】本発明の実施例による動きベクトル検出装置の構成を示すブロツク図である。
【図３】本発明の実施例による各動きベクトル値の説明に供する略線図である。
【図４】本発明の実施例による重み値を加えたブロツク差分値によるグラフを示す図表である。
【図５】他の実施例による動きベクトル検出装置の構成を示すブロツク図である。
【図６】画像上におけるブロツクマツチングの説明に供する略線図である。
【図７】連続パターンの画像においてブロツクマツチングを行つたときのＳ_umの最小値を表すグラフである。
【符号の説明】
１……動きベクトル検出装置、２……現フレーム用メモリ、３……前フレーム用メモリ、４……ブロツクポジシヨンカウント回路、５……基準ブロツクアドレスコントロール回路、６……上ブロツクベクトル値メモリ、７……左ブロツクベクトル値メモリ、８……差分演算回路、９……ベクトルコントロール回路、１０……評価ブロツクアドレスコントロール回路、１１、１２、３２……ベクトル差分重み演算回路、１３、１４……加算器、１５……ベクトル値検出回路、１６……係数演算回路、１７……乗算器、３１……前フレームベクトル値メモリ。[0001]
【table of contents】
The present invention will be described in the following order.
TECHNICAL FIELD OF THE INVENTION
Conventional technology
Problems to be Solved by the Invention (FIGS. 6 and 7)
Means for solving the problem
BEST MODE FOR CARRYING OUT THE INVENTION (FIGS. 1 to 5)
The invention's effect
[0002]
BACKGROUND OF THE INVENTION
The present invention relates to a motion vector detection device and a motion vector detection method, and is preferably applied to, for example, a case where a motion vector is detected when high-efficiency video is encoded in a digital HDTV (High Definition Television) VTR. Is.
[0003]
[Prior art]
Conventionally, when a moving image is encoded with high efficiency, there is a method using a motion vector, that is, a motion direction and size of an object in a temporally different image, for example, for motion compensation interframe encoding in high-efficiency encoding of an image. It is used. As a method for obtaining the motion vector of an image in this way, there is a blotting method (Japanese Patent Publication No. 54-124927).
[0004]
In this block clipping method, a block consisting of a suitable number of pixels (M × N pixels) is formed from the image of the current frame (basic image) and the image of the previous frame (reference image), which is one frame past. To do. Then, the block of the basic image is fixed (hereinafter, the block on the side to be fixed in this way is referred to as a reference block), and the block of the reference image is moved within the predetermined search area (hereinafter, this movement is performed). The block on the side is called an evaluation block), and the sum S of the differences between the pixel value of the reference block and the pixel value of the evaluation block S_umThe motion vector is detected by obtaining the position of the evaluation block on the reference image side where (x, y) is minimized.
[0005]
[Problems to be solved by the invention]
By the way, in the above-described blotching method, it is general that weighting by a vector is not originally performed. However, when weighting is performed on a zero vector position of a stationary part in an image, the zero vector position is the roof. When the entire screen moves (panning) in a block of continuous patterns such as roof tiles and block fences, or in a block where the image pattern with a constant brightness difference such as a blue sky or lawn has a constant brightness, the pattern continues. A motion vector detection error may occur.
However, when the weighted 0 vector position is in the still image area in the image, the motion vector is naturally 0 and detection errors hardly occur.
[0006]
That is, as shown in FIGS. 6 (A) and 6 (B), the entire screen moves from right to left when performing block clipping while paying attention to the block of a continuous image pattern such as a block (panning). In this case (only moving in the x direction, not moving in the y direction), there is a block similar to the image pattern of the reference block, so that the reference block is different from the original motion vector position in the search area. In some cases, the sum of the differences between the pixel value of and the pixel value of the evaluation block is minimized.
In practice, as shown in FIG. 7, the sum S of the differences between the pixel value of the reference block and the pixel value of the evaluation block_umSince there are a plurality (four places) of evaluation blocks at which (x, 0) is minimized at several positions in the search area, it is difficult to tell which evaluation block is genuine. There was a problem that a detection error occurred.
[0007]
The present invention has been made in consideration of the above points, and proposes a motion vector detection device and a motion vector detection method capable of reducing motion vector detection errors in an image having a continuous image pattern and a constant brightness image. It is something to try.
[0008]
[Means for Solving the Problems]
  In order to solve such a problem, in the present invention, in a motion vector detection device for detecting a motion vector of image data, image data of a reference block formed in a predetermined size and fixed at a predetermined position in the current frame is acquired. Reference block acquisition means, evaluation block acquisition means for acquiring image data of an evaluation block having the same size as the reference block at each position when moved within the search area of a predetermined range in the previous frame, and reference block acquisition means Difference calculating means for calculating a sum of differences of pixel values between the image data of the reference block acquired by the above and the image data of the evaluation block acquired by the evaluation block acquiring means, and the reference block of the reference block that has been detected. For the motion vector values for the surrounding block and the evaluation block for the reference block By multiplying the difference from the motion vector value by a coefficient proportional to the variance value of the image data of the reference block, the weight calculation means for calculating the weight amount and the weight amount calculated by the weight calculation means Addition calculation means for calculating the block difference value by adding to the sum of the pixel value differences calculated by the calculation means, and the evaluation block from the position where the block difference value calculated by the addition calculation means is minimum to the reference block Detection means for detecting a motion vector is provided.
[0009]
  In the motion vector detection device, the difference between the motion vector value for the surrounding block of the reference block for which the detection has been completed and the motion vector value for the evaluation block of the reference block is proportional to the variance value of the image data of the reference block. A weighting amount is calculated by multiplying the coefficient, and the weighting amount is added to the sum of the pixel value differences to detect a motion vector from the evaluation block at the position where the calculated block difference value is minimum to the reference block. Thus, even when there are a plurality of evaluation blocks that minimize the sum of pixel value differences, the final motion vector can be detected based on the weighting amount reflecting the variation tendency of the image data of the reference block. Therefore, even if it is a block image of a continuous pattern of the same pattern, it can detect a motion vector without error. It can be.
[0010]
  The surrounding block is the adjacent block closest to the reference block by being the upper block adjacent to the top of the reference block, the left block adjacent to the left of the reference block, or both the upper block and the left block. Since the weighting amount can be calculated based on the motion vector values up to the block, the calculation can be facilitated and the calculation time can be shortened, and thus the final motion vector can be detected in real time. .
[0011]
DETAILED DESCRIPTION OF THE INVENTION
Hereinafter, an embodiment of the present invention will be described in detail with reference to the drawings.
[0012]
As shown in FIG. 1A, a block for performing block clipping is formed by dividing a current frame (hereinafter referred to as a current frame) into a size of M × N pixels. The blocks divided in the current frame are (0, 0), (0, 1)... (0, n) in order from the upper left to the lower right.₀), (1, 0), (1, 1)... (2, 0), (2, 1)... (M, 0)... (M, n) For example, (m₀, N₀).
[0013]
Here, since the image data of the normal scanning line method (non-interlace method) is input from the upper left to the lower right, each block for detecting a motion vector also has (0, 0), (0, 1) …… (0, n₀), (1, 0), (1, 1) ... (2, 0), (2, 1) ... (m₀, N₀) ... The image data is processed in the order of (m, n).
In the present invention, the reference block (m₀, N₀) Of the surrounding blocks located around the reference block (m₀, N₀) The block adjacent to the top of the upper block (m₀-1, n₀), The block adjacent to the left is the left block (m₀, N₀-1).
[0014]
Further, as shown in FIG. 1B, the search area of the previous frame is a range (± x) having an area about twice the size of the reference block.₀, ± y₀) And the evaluation block (m_-1, N_-1) While moving the position forward, backward, left and right one pixel at a time, the reference block (m₀, N₀) And evaluation block (m_-1, N_-1).
[0015]
Next, FIG. 2 shows a configuration of a motion vector detection apparatus 1 for performing block clipping. In the motion vector detection device 1, the encoded image data D1 is sequentially stored in the current frame memory 2 and the previous frame memory 3 of the difference calculation unit 1A for each frame. In other words, the current frame memory 2 contains the current frame F.₀However, the previous frame F is stored in the previous frame memory 3._-1Will be stored respectively.
[0016]
A block position count circuit 4 detects the current frame F stored in the current frame memory 2 in order to detect a motion vector.₀Reference block fixed in (m₀, N₀) Is sent to the reference block address control circuit 5.
Here, since the image data is input once from the upper left to the lower right, the reference block (m₀, N₀) Around the reference block (m₀, N₀) Adjacent to the upper block (m₀-1, n₀) Or the left block adjacent to the left (m₀, N₀-1) is after the processing of the image data has already been performed, and the detection of the motion vector value is completed in each block.
[0017]
Accordingly, the block position count circuit 4 sends the designation signal S1 to the reference block address control circuit 5 and at the same time, the reference block (m₀, N₀) And the upper block (m₀-1, n₀) Motion vector value (x m-1, y n) (hereinafter referred to as the upper block motion vector value) and the left block (m₀, N₀-1) motion vector value (x m₀, Y n₀-1) (hereinafter referred to as the left block motion vector value) is sent to and stored in the upper block vector value memory 6 and the left block vector value memory 7, respectively.
[0018]
Here, as shown in FIG. 3A, the upper block motion vector value (x m-1, y n) is the previous frame F_-1From current upper block to current frame F₀Is a vector value up to the upper block, and is a fixed value already calculated. Also, the left block motion vector value (x m₀, Y n₀-1) is the previous frame F_-1From left block to current frame F₀Is a vector value up to the left block of, and is a fixed value already calculated.
[0019]
The reference block address control circuit 5 receives a reference block (m) from the current frame memory 2 based on the designation signal S1.₀, N₀) Is read out and sent to the current frame memory 2.
The current frame memory 2 uses the reference block (m₀, N₀) And is sent to the difference calculation circuit 8.
[0020]
Subsequently, as shown in FIG. 3B, the vector control circuit 9 performs the previous frame F in order to perform blocking._-1All evaluation blocks in the search area (m_-1, N_-1) To current frame F₀Standard block (m₀, N₀) In the order of motion vector values (x, y) (hereinafter referred to as reference block motion vector values), and an evaluation block address control circuit 10 and a vector difference weight calculation circuit 11 as motion vector value signals S3, 12 respectively.
[0021]
The evaluation block address control circuit 10 evaluates the evaluation block (m based on the input motion vector value signal S3._-1, N_-1) Is read out and sent to the previous frame memory 3. The previous frame memory 3 determines the evaluation block (m_-1, N_-1) And is sent to the difference calculation circuit 8.
[0022]
The difference calculation circuit 8 receives the input reference block (m₀, N₀) Image data D2 and evaluation block (m_-1, N_-1) Of the absolute value of the difference in pixel value from the image data D3 of S)_um(X, y) is calculated. As a result, the pixel value difference calculation unit 1 </ b> A makes the reference block (m₀, N₀) And evaluation block (m_-1, N_-1) And the pixel value difference can be calculated.
Here, the reference block (m₀, N₀) Where the address values of all the pixels in () are (h, v), the reference block (m₀, N₀) Is represented as Y (h, v) and the corresponding evaluation block (m_-1, N_-1) For each pixel value of Y When expressed as xy (h, v), the sum S of the absolute values of the differences between the pixel values_um(X, y) is the following formula
[Expression 1]

Can be expressed as
[0023]
On the other hand, the upper block vector value memory 6 of the block difference value calculation unit 1B stores the stored upper block motion vector value (x m₀-1, y n₀) To the vector difference weight calculation circuit 11. Similarly, the left block vector value memory 7 stores the stored left block motion vector value (x m₀, Y n₀-1) is sent to the vector difference weight calculation circuit 12.
[0024]
The vector difference weight calculation circuit 11 receives the input upper block motion vector value (x m₀-1, y n₀) And an evaluation block (m_-1, N_-1) To the reference block (m₀, N₀) To the reference block motion vector value (x, y) up to the difference (distance), the weight value W_upAs (x, y),
[Expression 2]

It is made to calculate with.
[0025]
Similarly, the vector difference weight calculation circuit 12 receives the input left block motion vector value (x m₀, Y n₀-1) and a value proportional to the difference (distance) between the reference block motion vector value (x, y) input from the vector control circuit 9 is a weight value W_leftAs (x, y),
[Equation 3]

It is made to calculate with.
[0026]
Here, the coefficient k used in the expressions (2) and (3) is a value that changes in accordance with the dispersion value (variation) of the image data D2 read from the current frame memory 2, and the coefficient k A new value is generated each time the position of the reference block is changed by the arithmetic circuit 16, and the weight value W_up(X, y) and W_left(X, y) is multiplied.
[0027]
That is, the coefficient calculation circuit 16 determines that each evaluation block (m_-1, N_-1) Sum of absolute values of differences from_umNaturally, (x, y) also has a large variation, so that the weight value W can be obtained by increasing the coefficient k._up(X, y) and W_left(X, y) is increased so as to be heavily weighted.
Further, when the variation of the image data D2 is small, the coefficient calculation circuit 16 determines each evaluation block (m_-1, N_-1) Sum of absolute values of differences from_umNaturally, (x, y) also has a small variation, so that the weight value W can be obtained by setting the coefficient k to a small value._up(X, y) and W_left(X, y) is made smaller and weighted smaller.
[0028]
As described above, the coefficient calculation circuit 16 varies the value of the coefficient k in accordance with the variation of the image data D2, and thereby the evaluation block (m_-1, N_-1) Sum S at each position_umWhen the values of (x, y) are weighted, they are not greatly changed (not canceled) by the coefficient k. Therefore, the weight value W_up(X, y) and W_left(X, y) is the sum S_umCompared with the value of (x, y), it becomes a very small value. In this case, the weight value W_up(X, y) and W_left(X, y) indicates that the reference block motion vector value (x, y) is the upper block motion vector value (x m₀-1, y n₀) And left block motion vector values (x m₀, Y n₀Since the distance is shorter as the distance is closer to -1), the difference is smaller and the weight value is naturally smaller.
[0029]
In this way, the calculated weight value W_up(X, y) and W_left(X, y) is added by the adder 13, and the sum S of the absolute values of the differences output from the difference calculation circuit 8 is further added._umBlock difference value D added to (x, y) via adder 14_iffAs (x, y),
[Expression 4]

And is sent to the vector value detection circuit 15 as detection means.
[0030]
In this way, all evaluation blocks (m_-1, N_-1Block difference value D calculated for_iff(X, y) is input to the vector value detection circuit 15. The vector value detection circuit 15 calculates all block difference values D_iff(X, y) are compared, and the block difference value D_iffThe reference block motion vector value (x, y) when (x, y) is the smallest is output.
The vector value detection circuit 15 also sends the detected reference block motion vector value (x, y) to the upper block vector value memory 6 and the left block vector value memory 7, and the right and lower blocks are then sent to the reference block ( m₀, N₀) Can be used as a motion vector value.
[0031]
In the above configuration, the motion vector detection device 1 has the reference block (m) shown in FIGS.₀, N₀) And evaluation block (m_-1, N_-1) And the case where the image moves from the right side to the left side.
In the motion vector detection device 1, the difference calculation unit 1A uses the current frame F.₀Standard block (m₀, N₀) Image data D2 and previous frame F_-1Evaluation block in the search area (m_-1, N_-1) Of the absolute value of the difference of each pixel value from the image data D3 of S)_um(X, y) is calculated by equation (1).
[0032]
Here, all evaluation blocks in the search area (m_-1, N_-1) Sum S_um(X, y) is calculated, and the evaluation block (m_-1, N_-1)), Since there are a plurality of block images of the same pattern of bricks, the sum S_umEvaluation block that minimizes (x, y) (m_-1, N_-1) (See FIG. 7), an accurate motion vector cannot be detected.
[0033]
Therefore, the motion vector detecting device 1 uses the upper block motion vector value (x) stored in advance by the block difference value calculation unit 1B. m₀-1, y n₀) And left block motion vector values (x m₀, Y n₀−1) and (FIG. 3A), the weight value W obtained by multiplying the absolute value of the difference between the reference block motion vector values (x, y) and (FIG. 3B) by the coefficient k._up(X, y), W_left(X, y) is calculated by the equations (2) and (3), and the weight value W_up(X, y), W_left(X, y) is the sum S of the absolute values of the differences between the pixel values according to equation (4)._umBlock difference value D added to (x, y)_iffFind (x, y).
[0034]
In practice, the motion vector detection apparatus 1 uses the upper block motion vector value (x m₀-1, y n₀) And the left block motion vector value (x m₀, Y n₀-1) is a fixed value, and all reference block motion vector values (x, y) in the search area are evaluated blocks (m_-1, N_-1), The evaluation value (m_-1, N_-1) Is a weight value W only when it is in a position representing the original motion vector in the search area._up(X, y), W_left(X, y) is minimized, and thus the block difference value D_iff(X, y) is minimized.
[0035]
Therefore, as shown in FIG. 4, the motion vector detecting apparatus 1 uses each evaluation block (m_-1, N_-1) Block difference value D at each position_iff(X, y) is the weight value W_up(X, y), W_leftIt is displayed as a graph having an inclination corresponding to the size of (x, y). As a result, the motion vector detecting device 1 is able to detect the block difference value D._iffThe reference block motion vector value (x, y) at the position where (x, y) is minimum can be detected by the vector value detection circuit 15, and thus a block image of a continuous pattern such as a brick can be obtained. Can detect the motion vector without error.
[0036]
In the motion vector detection device 1, the weight value W_up(X, y), W_leftWhen calculating (x, y), the value of the coefficient k is used as the reference block (m₀, N₀), The sum S_umBlock difference value D by weighting (x, y) by coefficient k without cancellation_iffThe angle of inclination of the graph representing (x, y) can be changed, thus the block difference value D_iffIt is possible to easily detect the reference block motion vector value (x, y) that minimizes (x, y), and reduce motion vector detection errors.
[0037]
In this way, the motion vector detection device 1 uses the difference calculation unit 1A to generate the reference block (m₀, N₀) And evaluation block (m_-1, N_-1) Sum of absolute values of differences in pixel values at respective positions_um(X, y) is calculated, and the weight value W is calculated by the block difference value calculation unit 1B._up(X, y), W_left(X, y) is calculated, and the weight value W_up(X, y), W_left(X, y) is evaluated block (m_-1, N_-1) Sum S at each position_umEach block difference value D added to (x, y)_iff(X, y) is calculated, and the block value difference D is detected by the vector value detection circuit 15 as detection means._iffA reference block motion vector value (x, y) that minimizes (x, y) can be detected.
[0038]
According to the above configuration, the motion vector detection device 1 has the reference block (m₀, N₀) Image data D2 and evaluation block (m_-1, N_-1) Of the absolute value of the difference of each pixel value from the image data D3 of S)_umThe upper block motion vector value (x m₀-1, y n₀) And left block motion vector values (x m₀, Y n₀-1) and the weight value W obtained by multiplying the absolute value of the difference between the reference block motion vector value (x, y) by the coefficient k_up(X, y), W_leftBlock difference value D by adding (x, y)_iffBy calculating (x, y), the block difference value D_iffThis makes it easy to detect the reference block motion vector value (x, y) that minimizes (x, y), and thus reduces the detection error of the motion vector even when the image patterns for each block are similar, resulting in high quality. Image data can be transmitted.
[0039]
In addition, the motion vector detection device 1 has a reference block (m₀, N₀) Is set to an upper block motion vector value (x m₀-1, y n₀) And left block motion vector values (x m₀, Y n₀-1) and the absolute value of the difference between the reference block motion vector value (x, y) and the sum S of the pixel value differences S_umWeight value W according to (x, y)_up(X, y), W_left(X, y) can be calculated, and thus the block difference value D_iffReference block when (x, y) is minimized (m₀, N₀) Motion vectors can be easily detected and detection errors can be reduced.
[0040]
Further, the motion vector detection device 1 is a reference block (m₀, N₀) Of the surrounding blocks located around the reference block (m₀, N₀) Adjacent to the upper block (m₀-1, n₀) And the left block (m₀, N₀-1) and the weight value W_up(X, y), W_leftBy calculating (x, y), the calculation can be facilitated and the calculation time can be shortened, and thus the motion vector can be detected in real time.
[0041]
In the above-described embodiment, the current vector F is detected in the motion vector detection device 1.₀Standard block (m₀, N₀) For the upper block (m₀-1, n₀) And left block (m₀, N₀-1) each motion vector value (x m₀-1, y n₀) And (x m₀, Y n₀−1) and the absolute value of the difference between the reference block motion vector value (x, y) and the weight value W_up(X, y), W_leftAlthough the case where (x, y) is calculated has been described, the present invention is not limited to this, and the previous frame in which the motion vector value of the previous frame is stored as in the motion vector detection device 30 shown in FIG. A vector value memory 31 and a vector difference weight calculation circuit 32 are added to the motion vector detection device 1, and based on the absolute value of the difference between the motion vector value of the previous frame and the reference block motion vector value (x, y). Weight value W_for(X, y) is calculated and the weight value W_for(X, y) is the weight value W_up(X, y), W_leftBlock difference value D in addition to (x, y)_iff(X, y) may be obtained. Also in this case, the same effect as the above-mentioned embodiment can be obtained.
[0042]
Further, in the above embodiment, the area range of the search area is set as the reference block (m₀, N₀However, the present invention is not limited to this, and the area range of the search area may be freely set according to the image pattern and processing time. good.
[0043]
Furthermore, in the above embodiment, the block difference value D_iffA reference block (m₀, N₀) Around the top and left of the upper block (m₀-1, n₀) And left block (m₀, N₀-1) each motion vector value (x m₀-1, y n₀) And (x m₀, Y n₀-1) has been described, but the present invention is not limited to this. In short, the reference block (m₀, N₀If the image data has already been processed and the motion vector value has been calculated, the block difference value D is determined by only the motion vector value of either the upper block or the left block._iff(X, y) may be obtained, or the motion vector value of the upper left block or the upper right block may be used.
[0044]
Further, in the above-described embodiments, the case where the present invention is applied to a VTR for HDTV has been described. However, the present invention is not limited to this, and a moving image is highly efficiently encoded and transmitted in the field of digital image communication. The present invention may be applied to various devices.
[0045]
【The invention's effect】
  As described above, according to the present invention, even when there are a plurality of evaluation blocks that minimize the total sum of pixel value differences, the final result is based on the weighting amount reflecting the variation tendency of the image data of the reference block. Since a motion vector can be detected, a motion vector can be detected without error even if it is a block image of a continuous pattern of the same pattern. Thus, in an image in which a similar image pattern is continuous and an image having a constant brightness. A motion vector detection device and a motion vector detection method capable of reducing motion vector detection errors can be realized.
[Brief description of the drawings]
BRIEF DESCRIPTION OF DRAWINGS FIG. 1 is a schematic diagram for explaining a case of performing block matching according to an embodiment of the present invention.
FIG. 2 is a block diagram showing a configuration of a motion vector detection device according to an embodiment of the present invention.
FIG. 3 is a schematic diagram for explaining each motion vector value according to the embodiment of the present invention.
FIG. 4 is a chart showing a graph of block difference values to which weight values are added according to an embodiment of the present invention.
FIG. 5 is a block diagram showing a configuration of a motion vector detection device according to another embodiment.
FIG. 6 is a schematic diagram for explanation of block clipping on an image.
FIG. 7 shows S when puncturing is performed on a continuous pattern image._umIt is a graph showing the minimum value of.
[Explanation of symbols]
DESCRIPTION OF SYMBOLS 1 ... Motion vector detection apparatus, 2 ... Current frame memory, 3 ... Previous frame memory, 4 ... Block position count circuit, 5 ... Reference block address control circuit, 6 ... Upper block vector value memory , 7... Left block vector value memory, 8... Difference calculation circuit, 9... Vector control circuit, 10... Evaluation block address control circuit, 11, 12, 32. ... Adder, 15... Vector value detection circuit, 16... Coefficient operation circuit, 17.

Claims (4)

In a motion vector detection device for detecting a motion vector of image data,
Reference block acquisition means for acquiring image data of a reference block formed in a predetermined size and fixed at a predetermined position in the current frame;
Evaluation block acquisition means for acquiring image data of an evaluation block having the same size as the reference block at each position when moved in a search area of a predetermined range in the previous frame;
Difference calculating means for calculating a sum of differences in pixel values between the image data of the reference block acquired by the reference block acquiring means and the image data of the evaluation block acquired by the evaluation block acquiring means;
A coefficient proportional to the variance value of the image data of the reference block is calculated with respect to the difference between the motion vector value for the surrounding block of the reference block for which detection has been completed and the motion vector value for the evaluation block of the reference block A weight calculation means for calculating a weighting amount by multiplication;
An addition calculation means for calculating a block difference value by adding the weighting amount calculated by the weight calculation means to a sum of differences of the pixel values calculated by the difference calculation means;
A motion vector detection apparatus comprising: a detection means for detecting a motion vector from the evaluation block to the reference block at a position where the block difference value calculated by the addition calculation means is minimum. 2. The surrounding block is an upper block adjacent to an upper portion of the reference block, a left block adjacent to a left portion of the reference block, or both the upper block and the left block. The motion vector detection device described. In a motion vector detection method for detecting a motion vector of image data,
A reference block acquisition step for acquiring image data of a reference block formed in a predetermined size and fixed at a predetermined position in the current frame;
An evaluation block acquisition step for acquiring image data of the evaluation block having the same size as the reference block at each position when moved in the search area of a predetermined range in the previous frame;
A difference calculation step for calculating a sum of differences in pixel values between the image data of the reference block acquired in the reference block acquisition step and the image data of the evaluation block acquired by the evaluation block acquisition means;
A coefficient proportional to the variance value of the image data of the reference block is calculated with respect to the difference between the motion vector value for the surrounding block of the reference block for which detection has been completed and the motion vector value for the evaluation block of the reference block A weight calculation step for calculating a weighting amount by multiplication;
An addition calculation step for calculating a block difference value by adding the weighting amount calculated in the weight calculation step to a sum of differences of the pixel values calculated in the difference calculation step;
A motion vector detection method comprising: a detection step for detecting a motion vector from the evaluation block to the reference block at a position where the block difference value calculated by the addition operation step is minimum. The peripheral block is block upper adjacent the top of the reference block, or left blocks adjacent to the left portion of the reference block, or to claim 3, characterized in that the both of the upper block and the left block The motion vector detection method described.

Images