CN1312924C

CN1312924C - Texture information based video image motion detecting method

Info

Publication number: CN1312924C
Application number: CNB2004100930986A
Authority: CN
Inventors: 董云朝; 邹琪; 郑世宝; 戈迪; 王峰
Original assignee: Shanghai Jiao Tong University
Current assignee: Shanghai Jiao Tong University
Priority date: 2004-12-16
Filing date: 2004-12-16
Publication date: 2007-04-25
Anticipated expiration: 2024-12-16
Also published as: CN1622613A

Abstract

The invention relates to a video image motion detection method based on texture information, which belongs to the field of electronic information. The present invention uses the inter-frame difference of the luminance signal and the pixel texture information in the neighborhood to match the two features to judge the movement of the pixel to be interpolated, and calculates the image at the same position in two adjacent fields separated by one frame period according to the input image. Then calculate the texture information of the above pixels, judge whether the texture information matches, and then judge the movement of the pixel to be interpolated according to the match between the frame difference and the texture information, and finally determine the pixel to be interpolated Motion detection results. While improving the ability to distinguish the two motion states of fast motion and stillness, the present invention also greatly improves the accuracy of the motion state, including the judgment of the motion direction, and can be used in LCOS, LCD, DLP, etc. that support progressive scanning. In the control device of various digital displays.

Description

Video Image Motion Detection Method Based on Texture Information

技术领域technical field

本发明涉及的是一种视频图象运动检测的方法，特别是一种基于纹理信息的视频图象运动检测的方法，属于电子信息领域。The invention relates to a video image motion detection method, in particular to a texture information-based video image motion detection method, which belongs to the field of electronic information.

背景技术Background technique

随着显示技术的不断发展和数字电视广播的逐渐推广，人们对画面的清晰度和逼真度的要求越来越高。传统电视广播由于带宽的限制采用了隔行扫描，每一幅画面上仅包含原始图象中一半的信息，分别构成奇场和偶场。隔行扫描不仅会造成图象垂直分辨率的下降，同时也存在着行间闪烁和大面积闪烁等缺陷。随着大尺寸高亮度的液晶电视机、等离子电视机进入家庭，隔行扫描方式造成的失真越来越让人无法忍受，为此高档电视机的芯片组中一般都配备了能完成去隔行功能的芯片。最简单的去隔行算法就是将属于同一帧的奇场和偶场图象直接拼插起来，形成一个输出帧。在图象静止的情况下，这样做是完全正确的。但是当奇场和偶场之间存在运动，尤其是水平方向的运动时，这种直接拼插的方法就会造成明显的锯齿，从而造成错误的输出图象，严重影响视觉效果。另外一种常用的去隔行算法为场内插值滤波方法。这种方法对运动画面的处理效果要优于直接拼插算法，但是对静止画面的处理就难以接受，为此人们开始研究在去隔行的过程中将这两种算法结合起来，根据图象的运动情况采用不同的方法，从而得到更好的去隔行效果。With the continuous development of display technology and the gradual promotion of digital TV broadcasting, people have higher and higher requirements for the clarity and fidelity of the picture. Due to the limitation of bandwidth, traditional TV broadcasting adopts interlaced scanning, and each picture only contains half of the information in the original image, which constitutes odd field and even field respectively. Interlaced scanning will not only cause a decrease in the vertical resolution of the image, but also have defects such as interline flicker and large-area flicker. As large-size and high-brightness LCD TVs and plasma TVs enter the family, the distortion caused by the interlaced scanning method is becoming more and more unbearable. Therefore, the chipsets of high-end TVs are generally equipped with a chip. The simplest de-interlacing algorithm is to directly interpolate the odd field and even field images belonging to the same frame to form an output frame. In the case of static images, this is completely correct. However, when there is movement between the odd field and the even field, especially the movement in the horizontal direction, this direct splicing method will cause obvious sawtooth, thereby causing wrong output images and seriously affecting the visual effect. Another commonly used deinterlacing algorithm is the intra-field interpolation filtering method. This method is better than the direct splicing and interpolation algorithm for moving pictures, but it is unacceptable for still pictures. For this reason, people have begun to study the combination of these two algorithms in the process of deinterlacing. Motion takes a different approach, resulting in better deinterlacing.

经文献检索发现，中国专利名称为：“检测交错视频序列中运动的方法和运动检测的装置”，申请号为：(01143659.X)。该专利提出了一种在视频信号处理系统中计算运动决定值的方法。根据该专利提供的方法，采用连续的三场图象数据计算帧差值来形成点状运动检测信号；从相邻两场的点状运动检测信号计算区状运动检测信号；从区状运动检测信号形成运动决定值，用于交错视频信号处理系统中的进一步处理。由于在计算帧差值的过程中仅采用了一组帧差信号，所以当图象中包含快速运动图象时，会造成运动状态的错判，即将快速运动判别为静止。After literature search, it was found that the Chinese patent name is: "Method for detecting motion in interlaced video sequence and device for motion detection", and the application number is: (01143659.X). This patent proposes a method for computing motion decision values in a video signal processing system. According to the method provided by this patent, the frame difference value is calculated by using continuous three-field image data to form a point motion detection signal; the zone motion detection signal is calculated from the point motion detection signals of two adjacent fields; the zone motion detection signal is obtained from the zone motion detection The signals form motion decision values for further processing in an interlaced video signal processing system. Since only one set of frame difference signals is used in the process of calculating the frame difference value, when the image contains fast moving images, it will cause a misjudgment of the motion state, that is, fast motion is judged as still.

发明内容Contents of the invention

本发明的目的在于克服现有技术中的不足，提供一种基于纹理信息匹配判断的视频图象运动检测的方法，用于去隔行模块或器件中的基于纹理信息的视频图象运动检测，使其能提高运动检测的准确性，进而提高去隔行变换的视觉效果，逼真地再现原始图象。The purpose of the present invention is to overcome the deficiencies in the prior art, provide a kind of method based on the video image motion detection of texture information matching judgment, be used for the video image motion detection based on texture information in deinterlacing module or device, make It improves the accuracy of motion detection, which in turn improves the visual effect of deinterlacing and faithfully reproduces the original image.

本发明是通过以下技术方案实现的，本发明采用亮度信号的帧间差异及其邻域内象素纹理信息匹配两个特征判断待插值象素的运动情况，根据输入图象计算间隔一个帧周期的相邻两场中处于相同位置的象素之间的帧间差异，然后计算以上象素的纹理信息，判断纹理信息是否匹配，再根据帧间差异和纹理信息匹配情况，判断待插值象素的运动情况，最终确定待插值象素的运动检测结果。The present invention is realized through the following technical solutions. The present invention uses the inter-frame difference of the luminance signal and the pixel texture information in the neighborhood to match the two features to judge the movement of the pixel to be interpolated, and calculates the motion of the pixel to be interpolated according to the input image. The inter-frame difference between the pixels at the same position in two adjacent fields, then calculate the texture information of the above pixels, judge whether the texture information matches, and then judge the pixel to be interpolated according to the inter-frame difference and the texture information matching situation Motion situation, and finally determine the motion detection result of the pixel to be interpolated.

以下对本发明作进一步的说明：The present invention is described further below:

(1)帧间差异的计算(1) Calculation of inter-frame differences

本发明计算两组帧间差异Diff_CFF和Diff_BF，计算方法如公式所示：The present invention calculates the differences Diff_CFF and Diff_BF between two groups of frames, and the calculation method is as shown in the formula:

$Diff Diff__CFF CFF [[m m]] = = \frac{11}{33} {Σ Σ}_{k k = = - - 11}^{11} | | I I ((i i,, j j + + m m + + k k,, n no)) - - I I ((i i,, j j - - m m + + k k,, n no - - 22)) | |$

$Diff Diff__BF BF [[m m]] = = \frac{11}{33} {Σ Σ}_{k k = = - - 11}^{11} | | I I ((i i + + 11,, j j + + m m + + k k,, n no + + 11)) - - I I ((i i + + 11,, j j - - m m + + k k,, n no - - 11)) | |$

其中I(i，j，n)表示第n场中第i行第j列象素的灰度，I(i，j，n-2)表示第n-2场中第i行第j列象素的灰度，I(i+1，j，n-1)表示第n-1场中第i+1行第j列象素的灰度，I(i+1，j，n+1)表示第n+1场中第i+1行第j列象素的灰度；m表示象素I(i，j，n)在水平方向的位移量，m∈[-N，N]，N表示象素I(i，j，n)的水平可检测运动范围；Diff_FFC[m]表示第n场与第n-2场间沿方向m的对应象素的差异，Diff_FB[m]表示第n+1场与第n-1场间沿方向m的对应象素的差异。Among them, I(i, j, n) represents the gray level of the i-th row and j-column pixel in the n-th field, and I(i, j, n-2) represents the i-th row and j-column image in the n-2 field. The gray level of the pixel, I(i+1, j, n-1) represents the gray level of the pixel in the i+1th row and the jth column in the n-1th field, I(i+1, j, n+1) Indicates the gray level of the pixel in the i+1th row and the jth column in the n+1th field; m represents the displacement of the pixel I(i, j, n) in the horizontal direction, m∈[-N, N], N Represents the horizontal detectable motion range of pixel I(i, j, n); Diff_FFC[m] represents the difference between the nth field and the n-2th field along the direction m of the corresponding pixel, Diff_FB[m] represents the nth The difference of the corresponding pixels along direction m between the +1 field and the n-1th field.

因为这两组差异涉及的输入场在时间上的间隔均为一个帧周期，所以把它们之间的差异叫做帧间差异。本发明检测的运动范围为(0，-N)到(0，N)之间，即当前象素I(i，j，n)在水平方向上[-N，N]内的运动情况。Because the time interval of the input fields involved in these two sets of differences is one frame period, the difference between them is called the inter-frame difference. The motion range detected by the present invention is between (0,-N) and (0, N), that is, the motion situation of the current pixel I(i, j, n) in [-N, N] in the horizontal direction.

(2)纹理信息匹配的判断(2) Judgment of texture information matching

纹理信息匹配判断包含纹理信息的计算和纹理信息匹配的判断。与帧间差异相对应，纹理信息的匹配也分为第n-2场与第n场之间的匹配和第n-1场与第n+1场之间的匹配，分别以Accordance_CFF[m]和Accordance_BF[m]表示。这里以Accordance_BF[m]为例说明纹理信息匹配判断的过程。假如第n-1场的象素I(i，j-m，n-1)沿方向m运动经过第n场的待插值象素I(i，j+1，n)到第n+1场的象素I(i，j+m，n+1)，则需要考虑以下两个集合内的象素变化，即以下区域内的象素纹理特征是否一致：{I(i，j+m+2(k-2)，n+1)，k∈[0，4]}和{I(i，j-m+2(k-2)，n-1)，k∈[0，4]}。这一区域内的纹理信息按以下公式所示收集：The texture information matching judgment includes the calculation of texture information and the judgment of texture information matching. Corresponding to the difference between frames, the matching of texture information is also divided into the matching between the n-2th field and the nth field and the matching between the n-1th field and the n+1th field, respectively with Accordance_CFF[m] and Accordance_BF[m] said. Here, Accordance_BF[m] is taken as an example to illustrate the process of texture information matching judgment. If the pixel I(i, j-m, n-1) of the n-1th field moves along the direction m through the interpolation pixel I(i, j+1, n) of the nth field to the image of the n+1th field Pixel I(i, j+m, n+1), you need to consider the pixel changes in the following two sets, that is, whether the pixel texture features in the following areas are consistent: {I(i, j+m+2( k-2), n+1), k ∈ [0, 4]} and {I(i, j-m+2(k-2), n-1), k ∈ [0, 4]}. Texture information in this area is gathered as shown in the following formula:

$\{\begin{matrix} GradBw GradBw ((m m,, k k,, n no + + 11)) = = I I ((i i,, j j + + m m + + 22 ((k k - - 11)),, n no + + 11)) - - I I ((i i,, j j + + m m + + 22 k k,, n no + + 11)) k k &Element; &Element; [[0,1 0,1]] \\ GradFw GradFw ((i i,, j j,, n no - - 11)) = = I I ((i i,, j j - - m m + + 22 ((k k - - 11)),, n no - - 11)) - - I I ((i i,, j j - - m m + + 22 k k,, n no - - 11)) k k &Element; &Element; [[0,1 0,1]] \end{matrix}$

$\{\begin{matrix} GradBw GradBw__Border Border ((m m,, 00,, n no + + 11)) = = I I ((i i,, j j + + m m - - 44,, n no + + 11)) - - I I ((i i,, j j + + m m - - 22,, n no + + 11)) \\ GradBw GradBw__Border Border ((m m,, 11,, n no + + 11)) = = I I ((i i,, j j + + m m + + 44,, n no + + 11)) - - I I ((i i,, j j + + m m + + 22,, n no + + 11)) \\ GradFw GradFw__Border Border ((m m,, 00,, n no - - 11)) = = I I ((i i,, j j - - m m - - 44,, n no - - 11)) - - I I ((i i,, j j - - m m - - 22,, n no - - 11)) \\ GradFw GradFw__Border Border ((m m,, 00,, n no - - 11)) = = I I ((i i,, j j - - m m + + 44,, n no - - 11)) - - I I ((i i,, j j - - m m + + 22,, n no - - 11)) \end{matrix}$

其中GradBw和GradFw表示中心区域的纹理信息，GradBw_Border和GradFw_Border表示边界区域的纹理信息。Among them, GradBw and GradFw represent the texture information of the central area, and GradBw_Border and GradFw_Border represent the texture information of the border area.

纹理信息匹配的判断如下：The judgment of texture information matching is as follows:

①如果GradBw与GradFw中的元素全部小于某一门限，则转到③；否则顺序执行；① If all the elements in GradBw and GradFw are less than a certain threshold, then go to ③; otherwise, execute sequentially;

②如果GradBw与GradFw中的对应元素符号一致，则转到③；否则设置匹配标志参数AccordanceBF[m]＝0，纹理匹配判断结束；② If the corresponding element symbols in GradBw and GradFw are consistent, then go to ③; otherwise set the matching flag parameter AccordanceBF[m]=0, and the texture matching judgment ends;

③如果GradBw_Border与GradFw_Border中的元素全部小于某一门限，则设置匹配标志参数AccordanceBF[m]＝1，纹理匹配判断结束；否则顺序执行；③ If the elements in GradBw_Border and GradFw_Border are all smaller than a certain threshold, then set the matching flag parameter AccordanceBF[m]=1, and the texture matching judgment ends; otherwise, execute sequentially;

④如果GradBw_Border与GradFw_Border中的元素符号一致，则设置相应的匹配标志参数AccordanceBF[m]＝1，纹理匹配判断结束；否则顺序执行；④ If the element symbols in GradBw_Border and GradFw_Border are consistent, then set the corresponding matching flag parameter AccordanceBF[m]=1, and the texture matching judgment ends; otherwise, execute sequentially;

⑤设置匹配标志参数AccordanceBF[m]＝0，纹理匹配判断结束。⑤ Set the matching flag parameter AccordanceBF[m]=0, and the texture matching judgment ends.

(3)确定运动检测结果(3) Determine the motion detection result

确定运动检测结果包括运动系数的确定和运动方向的确定，象素的运动系数用Motion表示，运动方向用Dir表示。过程如下：Determining the motion detection result includes determining the motion coefficient and the motion direction. The motion coefficient of the pixel is represented by Motion, and the motion direction is represented by Dir. The process is as follows:

①假设Diff_CFF[m1]＝min(Diff_CFF[m]，m∈[-N，N])，Diff_BF[m2]＝min(Diff_BF[m]，m∈[-N，N])，则m1和m2分别称为Diff_CFF和Diff_BF的相关运动方向。① Suppose Diff_CFF[m1]=min(Diff_CFF[m], m∈[-N, N]), Diff_BF[m2]=min(Diff_BF[m], m∈[-N, N]), then m1 and m2 The associated motion directions are called Diff_CFF and Diff_BF, respectively.

②如果Diff_CFF[m1]和Diff_BF[m1]都小于规定的门限，并且Accordance_CFF[m1]和Accordance_BF[m1]都为1，则Motion＝0，并且Dir＝m1。② If both Diff_CFF[m1] and Diff_BF[m1] are smaller than the specified threshold, and both Accordance_CFF[m1] and Accordance_BF[m1] are 1, then Motion=0, and Dir=m1.

③如果Diff_CFF[m2]和Diff_BF[m2]都小于规定的门限，并且Accordance_CFF[m2]和Accordance_BF[m2]都为1，则Motion＝0，并且Dir＝m2。③ If both Diff_CFF[m2] and Diff_BF[m2] are smaller than the specified threshold, and both Accordance_CFF[m2] and Accordance_BF[m2] are 1, then Motion=0, and Dir=m2.

④否则确定Motion＝1，同时Dir＝0。④ Otherwise, it is determined that Motion=1 and Dir=0 at the same time.

本发明具有实质性特点和显著进步，本发明用于完成对隔行扫描视频图象的运动检测，使其可以应用到去隔行模块或器件中。本发明首先根据相邻的四场输入视频图象计算两组帧间差异，并判断运动检测区域内象素纹理特征的匹配情况；根据两组帧间差异和象素纹理特征的匹配情况确定运动检测结果。在提高对快速运动和静止这两种运动状态的区分能力的同时，也大大提高了运动状态，包括运动方向的判断的准确性。可以用在支持逐行扫描的LCOS、LCD、DLP等多种数字显示器的控制器件中。The invention has substantive features and remarkable progress, and the invention is used to complete the motion detection of interlaced scanning video images, so that it can be applied to deinterlacing modules or devices. The present invention first calculates the difference between two groups of frames according to the adjacent four-field input video images, and judges the matching situation of the pixel texture feature in the motion detection area; determines the motion according to the matching situation of the difference between the two groups of frames and the pixel texture feature Test results. While improving the ability to distinguish the two motion states of fast motion and rest, it also greatly improves the accuracy of the motion state, including the judgment of the motion direction. It can be used in control devices of various digital displays such as LCOS, LCD, and DLP that support progressive scanning.

附图说明Description of drawings

图1本发明所涉及的运动状态判断原理框图。FIG. 1 is a block diagram of the motion state judgment principle involved in the present invention.

图2本发明所涉及的帧间差异计算示意图。Fig. 2 is a schematic diagram of inter-frame difference calculation involved in the present invention.

图3本发明所涉及的纹理信息计算示意图。Fig. 3 is a schematic diagram of texture information calculation involved in the present invention.

具体实施方式Detailed ways

结合本发明的内容提供以下实施例：Provide following embodiment in conjunction with content of the present invention:

本发明采用亮度信号的帧间差异及其邻域内象素纹理信息的差异作为运动检测的判断依据。对当前象素在水平方向[-N，+N]范围内的运动进行检测，并给出帧间运动插值的系数。The present invention adopts the inter-frame difference of the brightness signal and the difference of the texture information of the pixels in the neighborhood as the judgment basis of the motion detection. Detect the motion of the current pixel in the range [-N, +N] in the horizontal direction, and give the inter-frame motion interpolation coefficient.

对隔行扫描的输入序列，假设第n场中包含原始图象中的第i行，第i+2行，第i+4行等，去隔行的目的是尽可能逼真地补充第n场中缺少的第i+1行，第i+3行，第i+5行等。如果在第n场和第n-1场之间没有运动，则第n场中缺少的信息将直接用第n-1场中的信息补充，否则将根据两场之间的运动情况在场内插值和帧间插值之间加权求和得到。由此可见对第n场和第n-1场之间的运动情况进行准确的判断是得到正确的去隔行结果的关键。For an interlaced input sequence, it is assumed that field n contains line i, line i+2, line i+4, etc. in the original image. The purpose of deinterlacing is to supplement the missing data in field n as realistically as possible. Line i+1, line i+3, line i+5, etc. If there is no motion between field n and field n-1, the missing information in field n will be directly supplemented with information in field n-1, otherwise it will be interpolated within the field based on the motion between the two fields and the weighted summation between frame interpolation. It can be seen that the key to obtaining a correct de-interlacing result is to accurately judge the motion between the nth field and the n-1th field.

如图1所示，是本发明原理框图。运动检测过程如下：对隔行扫描的视频图象顺序经过三个场延迟模块10，11和12，分别得到第n+1场的象素集合{I(i，j+m，n+1)，m∈[-N，N]}21，第n场的象素集合{I(i，j+m，n)，m∈[-N，N]}22，第n-1场的象素集合{I(i，j+m，n-1)，m∈[-N，N]}23，以及第n-2场的象素集合{I(i，j+m，n-2)，m∈[-N，N]}24。象素集合22和24进入帧间差异计算I模块13，得到帧间差异集合{Diff_CFF[m]，m∈[-N，N]}，用信号31表示；22和24进入纹理匹配判断I模块14，得到纹理匹配结果集合{Accordance_CFF[m]，m∈[-N，N]}，用信号32表示。象素集合21和23进入帧间差异计算II模块15，得到帧间差异集合{Diff_BF[m]，m∈[-N，N]}，用信号33表示；21和23进入纹理匹配判断II模块15，得到纹理匹配结果集合{Accordance_BF[m]，m∈[-N，N]}，用信号34表示。信号31、32、33、34送入运动检测模块17，完成最后的运动状态检测，得到运动检测系数Motion，用信号41表示，和运动方向Dir，用信号42表示。As shown in Fig. 1, it is a functional block diagram of the present invention. The motion detection process is as follows: through three field delay modules 10, 11 and 12 to the interlaced video image order, obtain the pixel set {I(i, j+m, n+1) of the n+1 field respectively, m∈[-N, N]}21, the pixel set of the nth field {I(i, j+m, n), m∈[-N, N]}22, the pixel set of the n-1th field {I(i, j+m, n-1), m∈[-N, N]}23, and the pixel set {I(i, j+m, n-2), m of the n-2th field ∈[-N,N]}24. Pixel sets 22 and 24 enter the inter-frame difference calculation I module 13 to obtain the inter-frame difference set {Diff_CFF[m], m∈[-N, N]}, represented by signal 31; 22 and 24 enter the texture matching judgment I module 14. Obtain a texture matching result set {Accordance_CFF[m], m∈[-N, N]}, represented by a signal 32 . Pixel sets 21 and 23 enter the inter-frame difference calculation II module 15 to obtain the inter-frame difference set {Diff_BF[m], m∈[-N, N]}, represented by signal 33; 21 and 23 enter the texture matching judgment II module 15. Obtain a texture matching result set {Accordance_BF[m], m∈[-N, N]}, represented by a signal 34 . The signals 31, 32, 33, 34 are sent to the motion detection module 17 to complete the final motion state detection to obtain the motion detection coefficient Motion, represented by the signal 41, and the motion direction Dir, represented by the signal 42.

图2所示，为本发明所涉及的帧间差异计算的示例。假设第n-1场中象素I(i+1，j+1，n-1)水平移动-1个象素到达第n场中(i+1，j)处，并沿此方向到达第n+1场中的(i+1，j-1)处。可见当m＝-1时，I(i+1，j+1，n-1)与I(i+1，j-1，n+1)是同一运动轨迹上的两个象素。考虑到图象的邻域相关性，在计算帧间差异时，不仅考虑运动轨迹上的象素，还考虑与这些象素左右相邻的象素。用公式(1)计算它们的帧间差异。As shown in FIG. 2, it is an example of inter-frame difference calculation involved in the present invention. Assume that pixel I(i+1, j+1, n-1) in the n-1th field moves horizontally by -1 pixel to reach (i+1, j) in the nth field, and arrives at the (i+1, j-1) in the n+1 field. It can be seen that when m=-1, I(i+1, j+1, n-1) and I(i+1, j-1, n+1) are two pixels on the same motion track. Considering the neighborhood correlation of the image, when calculating the inter-frame difference, not only consider the pixels on the motion track, but also consider the pixels adjacent to these pixels left and right. Calculate their inter-frame differences using Equation (1).

图3所示，为本发明所涉及的纹理信息计算的示例。假设第n-1场中象素I(i+1，j+1，n-1)水平移动-1个象素到达第n场中(i+1，j)处，并沿此方向到达第n+1场中的(i+1，j-1)处。可见当m＝-1时，I(i+1，j+1，n-1)与I(i+1，j-1，n+1)是同一运动轨迹上的两个象素。考虑到图象的邻域相关性，并且为了提高纹理判断模块的抗噪性能，在计算纹理信息时采用的象素除了运动轨迹上的象素外，还包含与这些象素左右间隔一个象素的象素。用公式(2)和(3)计算纹理信息。As shown in FIG. 3, it is an example of texture information calculation involved in the present invention. Assume that pixel I(i+1, j+1, n-1) in the n-1th field moves horizontally by -1 pixel to reach (i+1, j) in the nth field, and arrives at the (i+1, j-1) in the n+1 field. It can be seen that when m=-1, I(i+1, j+1, n-1) and I(i+1, j-1, n+1) are two pixels on the same motion track. Considering the neighborhood correlation of the image, and in order to improve the anti-noise performance of the texture judgment module, the pixels used in the calculation of the texture information include not only the pixels on the motion track, but also a pixel spaced from the left and right of these pixels. of pixels. The texture information is calculated using formulas (2) and (3).

首先计算帧间差异和纹理信息。特例中检测的象素运动范围N为2，即检测待插值象素I(i+1，j，n)在水平方向上[-2，2]范围内的运动情况。根据公式(1)计算帧间差异{Diff_CFF[m]，m∈[-2，2]}和{Diff_BF[m]，m∈[-2，2]}，在特例中用15做为帧间差异的门限。根据公式(2)和(3)计算纹理信息，以10做为纹理信息匹配判断的门限，判断纹理信息是否匹配，得到集合{Accordance_CFF[m]，m∈[-2，2]}和{Accordance_BF[m]，m∈[-2，2]}。最后根据帧间差异和纹理匹配判断结果确定待插值象素I(i+1，j，n)的运动检测系数及运动方向。The inter-frame differences and texture information are first computed. In the special case, the detected pixel motion range N is 2, that is, to detect the motion of the pixel to be interpolated I(i+1, j, n) in the range [-2, 2] in the horizontal direction. Calculate the inter-frame difference {Diff_CFF[m], m∈[-2, 2]} and {Diff_BF[m], m∈[-2, 2]} according to formula (1), and use 15 as the inter-frame in a special case The difference threshold. Calculate the texture information according to formulas (2) and (3), use 10 as the threshold for texture information matching judgment, and judge whether the texture information matches, and get the set {Accordance_CFF[m], m∈[-2, 2]} and {Accordance_BF [m], m ∈ [-2, 2]}. Finally, determine the motion detection coefficient and motion direction of the interpolated pixel I(i+1, j, n) according to the inter-frame difference and the texture matching judgment result.

采用以上方法对快速水平运动(水平运动速度为6象素/场)、静止图象和慢速水平运动图象(水平运动速度为2象素/场)等三组视频序列(每组各10帧)进行了运动检测试验。结果表明，本专利对以上三种运动状态判断的正确率均达到95％以上，对静止图象的运动状态判断达到100％。Adopt above method to fast horizontal movement (horizontal movement speed is 6 pixels/field), static image and slow horizontal movement picture (horizontal movement speed is 2 pixels/field) etc. three groups of video sequences (every group respectively 10 frame) for motion detection experiments. The results show that the correct rate of the patent for judging the above three kinds of motion states all reaches more than 95%, and the judgment of the motion state of still images reaches 100%.

Claims

1, a kind of video image motion detecting method based on texture information, it is characterized in that, according to the frame difference between the pixel that is in same position among adjacent two an of frame period of input imagery counting period, calculate the texture information of above pixel then, judge whether texture information mates, according to frame difference and texture information match condition, judge the motion conditions of interpolation pixel again, finally determine the motion detection result of interpolation pixel.

2, the video image motion detecting method based on texture information according to claim 1 is characterized in that, the computing formula of described frame difference is as follows:

Diff_CFF [m] = \frac{1}{3} Σ_{k = - 1}^{1} | I (i, j + m + k, n) - I (i, j - m + k, n - 2) |

Diff_BF [m] = \frac{1}{3} Σ_{k = - 1}^{1} | I (i + 1, j + m + k, n + 1) - I (i + 1, j - m + k, n - 1) |

Wherein I represents the luma component values of pixel, and n is a sequence number, i, j is respectively the row number of a middle pixel and row number, m is five horizontal detection direction numberings, and span is [2 ,-1,0,1,2], k is the horizontal displacement value of pixel, its scope be from zero to the images the pixel number of delegation, Diff_CFF[m] pixel I (i, j+m in the expression n field, n) and horizontal direction neighborhood [1,1] Nei pixel and the pixel I in the n-2 field (i, j-m, n-2) and horizontal direction neighborhood [1,1] frame difference between Nei the pixel, Diff_BF[m] expression when motion vector be (0, in the time of 2m), the pixel I (i+1 in the n+1 field, j+m, n+1) and horizontal direction neighborhood [1,1] in pixel and the pixel I in the n-1 field (i+1, j-m, n-1) and horizontal direction neighborhood [1,1] in pixel between frame difference.

3, the video image motion detecting method based on texture information according to claim 1 and 2 is characterized in that, whether described texture information mates, and comprises the calculating of texture information and carries out the judgement of texture information coupling according to result of calculation:

Following formula is adopted in the calculating of described texture information:

[\begin{matrix} GradBw (m, k, n + 1) = I (i, j + m + 2 (k - 1), n + 1) - I (i, j + m + 2 k, n + 1) & k &Element; [0, 1] \\ GradFw (i, j, n - 1) = I (i, j - m + 2 (k - 1), n - 1) - I (i, j - m + 2 k, n - 1) & k &Element; [0, 1] \end{matrix}]

\{\begin{matrix} GradBw_Border (m, 0, n + 1) = I (i, j + m - 4, n + 1) - I (i, j + m - 2, n + 1) \\ GradBw_Border (m, 1, n + 1) = I (i, j + m + 4, n + 1) - I (i, j + m + 2, n + 1) \\ GradFw_Border (m, 0, n - 1) = I (i, j - m - 4, n - 1) - I (i, j - m - 2, n - 1) \\ GradFw_Border (m, 0, n - 1) = I (i, j - m + 4, n - 1) - I (i, j - m + 2, n - 1) \end{matrix}

Wherein GradBw and GradFw represent the texture information of central area, and GradBw_Border and GradFw_Border represent the texture information of borderline region;

Described texture information matching judgment is as follows:

If 1. 3. the element among GradBw and the GradFw then forwards to all less than a certain thresholding; Otherwise order is carried out;

If 2. GradBw is consistent with the corresponding element symbol among the GradFw, then forward to 3.; Otherwise match flag parameter A ccordanceBF[m is set]=0, the texture matching judgment finishes;

If 3. the element among GradBw_Border and the GradFw_Border is all less than a certain thresholding, match flag parameter A ccordanceBF[m is set then]=1, the texture matching judgment finishes; Otherwise order is carried out;

If 4. GradBw_Border is consistent with the symbol of element among the GradFw_Border, corresponding match flag parameter A ccordanceBF[m is set then]=1, the texture matching judgment finishes; Otherwise order is carried out;

5. match flag parameter A ccordanceBF[m is set]=0, the texture matching judgment finishes.

4, the video image motion detecting method based on texture information according to claim 3 is characterized in that, described definite motion detection result, comprise determining of kinematic coefficient and determining of the direction of motion, the kinematic coefficient of pixel represents that with Motion the direction of motion represents that with Dir process is as follows:

1. suppose Diff_CFF[m1]=min (Diff_CFF[m], m ∈ [N, N]), Diff_BF[m2]=min (Diff_BF[m], m ∈ [N, N]), then m1 and m2 are called the relevant direction of motion of Diff_CFF and Diff_BF;

If 2. Diff_CFF[m1] and Diff_BF[m1] all less than the thresholding of regulation, and Accordance_CFF[m1] and Accordance_BF[m1] all be 1, then Motion=0, and Dir=m1;

If 3. Diff_CFF[m2] and Diff_BF[m2] all less than the thresholding of regulation, and Accordance_CFF[m2] and Accordance_BF[m2] all be 1, then Motion=0, and Dir=m2;

4. otherwise determine Motion=1, Dir=0.