CN102063699A

CN102063699A - A Method of Video Image Scaling Based on Interpolation Orthogonal Multiwavelet

Info

Publication number: CN102063699A
Application number: CN 201010249887
Authority: CN
Inventors: 罗笑南; 蔡琼; 王栋; 殷伟; 李苗; 林格
Original assignee: Zhongshan Dingyu Electronic Technology Co ltd; Dongguan Huanya Hi Tech Electronics Co ltd; Sun Yat Sen University
Current assignee: Zhongshan Dingyu Electronic Technology Co ltd; Dongguan Huanya Hi Tech Electronics Co ltd; Sun Yat Sen University
Priority date: 2010-08-10
Filing date: 2010-08-10
Publication date: 2011-05-18

Abstract

The embodiment of the invention discloses a video image scaling method based on interpolation orthogonal multi-wavelets, which comprises the following steps: acquiring original image data; performing orthogonal multi-wavelet decomposition on original image data; performing interpolation processing on the wavelet coefficient subjected to wavelet decomposition; reconstructing the orthogonal multi-wavelet subjected to interpolation processing to obtain a zoomed image; and outputting the scaled image. Because the embodiment of the invention adopts the video image scaling method based on the interpolation orthogonal multi-wavelet, the edge of the image can be enhanced, the edge blur of the scaled image is overcome, and the contrast and the definition of the image are improved.

Description

A Method of Video Image Scaling Based on Interpolation Orthogonal Multiwavelet

技术领域technical field

本发明涉及数字图像处理的技术领域，具体涉及一种基于插值正交多小波的视频图像缩放的方法。The invention relates to the technical field of digital image processing, in particular to a video image scaling method based on interpolation orthogonal multi-wavelet.

背景技术Background technique

数字视频处理问题相对于数字图像处理的一个显著特点就是对实时性的要求很高。通常对于人的眼睛，18帧/秒以下的动态画面在观看时会觉得不流畅，在相同分辨率下，帧数越高的记录方式，人眼的观赏感觉肯定是更舒适，一般来说，达到每秒60帧记录的时候，人眼的观赏舒适度将达到饱和的顶点状态。通常的CIF(720.576像素)格式，每秒30帧的视频数据像素周期为74ns。很多情况下，视频实际分辨率大小与显示设备分辨率大小不一致，这时候就需要对视频进行拉伸或缩小来产生目标大小的画面，而改变分辨率的过程是通过对视频的每一帧画面进行缩放来完成的。A notable feature of digital video processing compared to digital image processing is the high requirement for real-time performance. Usually, for human eyes, dynamic pictures below 18 frames per second will feel unsmooth when watching. At the same resolution, the higher the number of frames, the more comfortable the viewing experience for human eyes. Generally speaking, When recording at 60 frames per second, the viewing comfort of human eyes will reach the peak state of saturation. In the usual CIF (720.576 pixels) format, the video data pixel period of 30 frames per second is 74ns. In many cases, the actual resolution of the video is inconsistent with the resolution of the display device. At this time, the video needs to be stretched or shrunk to produce a picture of the target size. The process of changing the resolution is to change each frame of the video Scaling is done.

数字图像的缩放通常借助图像插值来实现。插值算法的好坏将直接关系到图像的失真程度。用图像插值算法进行图像缩放时，通常存在一对相悖的要素：图像处理速度和图像精度。通常情况下要获得高速甚至实时的图像输出，只能采用相对简单、运算量小的插值算法；而要获得高精度的处理结果，只能牺牲速度，采用复杂度高的算法。插值一般满足两个条件：插值的像素值在二维空间是一个连续曲面以及在原图的采样点上插值结果等于图的像素值。图像线性插值可以看作是原始图像经过一个线性滤波器或者线性插值函数的输出。插值过程是首先把插值图像的像素坐标映射到原始图像坐标中，映射后的坐标值通常都不是整数，然后通过线性插值方法，根据该坐标周围的像素值得到该像素的狄度值。线性插值方法主要利用已知的像素点的加权平均，也就是图像的离散采样值和插值函数之间的二维卷积来计算未知像素点的值，选取不同的插值方法和插值函数可以得到不同效果的重建图像。The scaling of digital images is usually realized by means of image interpolation. The quality of the interpolation algorithm will be directly related to the degree of image distortion. When using image interpolation algorithm for image scaling, there is usually a pair of contradictory elements: image processing speed and image accuracy. Usually, to obtain high-speed or even real-time image output, only a relatively simple interpolation algorithm with a small amount of calculation can be used; while to obtain high-precision processing results, one can only sacrifice speed and use a highly complex algorithm. Interpolation generally satisfies two conditions: the interpolated pixel value is a continuous surface in two-dimensional space and the interpolation result at the sampling point of the original image is equal to the pixel value of the image. Image linear interpolation can be regarded as the output of the original image through a linear filter or linear interpolation function. The interpolation process is to first map the pixel coordinates of the interpolated image to the original image coordinates. The mapped coordinate values are usually not integers, and then use the linear interpolation method to obtain the Dido value of the pixel according to the pixel values around the coordinates. The linear interpolation method mainly uses the weighted average of known pixels, that is, the two-dimensional convolution between the discrete sampling value of the image and the interpolation function to calculate the value of the unknown pixel. Different interpolation methods and interpolation functions can be selected to obtain different values. The reconstructed image of the effect.

在对此方法的研究和实践过程中，本发明的发明人发现尽管这种内插算法简单易实现，且在局部可以得到很高的精度，但在缩小的过程中由于其局部性，没考虑到图像在全局的结构特征，所以有可能失去原图像大的结构信息，使图像的一些关键的边缘特征有可能丢失；在放大的过程中由于仅是低频部分的分解，没有考虑高频信息，所以有可能失去原图像的细节信息，使放大的图像很模糊，随着放大倍数的增加，图像的清晰度明显降低，出现锯齿边缘效应。During the research and practice of this method, the inventors of the present invention found that although this interpolation algorithm is simple and easy to implement, and can obtain high local accuracy, it is not considered in the process of shrinking due to its locality. Because of the global structural features of the image, it is possible to lose the large structural information of the original image, and some key edge features of the image may be lost; in the process of enlarging, because only the low-frequency part is decomposed, high-frequency information is not considered. Therefore, it is possible to lose the detail information of the original image and make the enlarged image blurred. With the increase of magnification, the sharpness of the image is obviously reduced, and the jagged edge effect appears.

发明内容Contents of the invention

本发明提供一种基于插值正交多小波的视频图像缩放方法，该方法能够增强图像的边缘，克服了分形插值引起的图像边缘模糊，从而提高了图像的对比度、清晰度。放大后图像轮廓光滑清晰，图像内部的对比度有所增强。The invention provides a video image scaling method based on interpolation orthogonal multi-wavelet, the method can enhance the edge of the image, overcome the blurred image edge caused by fractal interpolation, thereby improving the contrast and clarity of the image. After zooming in, the outline of the image is smooth and clear, and the contrast inside the image is enhanced.

本发明实施例提供一种基于插值正交多小波的视频图像缩放方法，包括：An embodiment of the present invention provides a video image scaling method based on interpolation orthogonal multi-wavelet, including:

获取原始图像数据；Get raw image data;

对原始图像数据进行正交多小波分解；Orthogonal multi-wavelet decomposition is performed on the original image data;

对进行小波分解后的小波系数进行插值处理；Perform interpolation processing on the wavelet coefficients after wavelet decomposition;

将进行插值处理后的正交多小波重建得到缩放的图像；Reconstruct the interpolated orthogonal multi-wavelet to obtain a scaled image;

将缩放后的图像输出。Output the scaled image.

所述对原始图像数据进行正交多小波分解包括：Carrying out the orthogonal multi-wavelet decomposition on the original image data includes:

首先：对图像的行和列进行向量化，其中：一维离散信号的向量化如下：First: vectorize the rows and columns of the image, where: the vectorization of a one-dimensional discrete signal is as follows:

$&ForAll; f (t) &Element; V_{0},$ 则有 $&ForAll; f (t) &Element; V_{0},$ then there is

根据多尺度函数与多小波的插值性质可得：According to the interpolation properties of multi-scale function and multi-wavelet:

C_0，n＝(c_0，0，n，c_1，0，n)^T＝(f(n)，f(n+1/2))^T C _0,n =(c _0,0,n ,c _1,0,n ) ^T =(f(n),f(n+1/2)) ^T

其中

in

如果把f(x)在整数点和半整数点上的抽样看成离散信号，即x(n)＝f(n/2)，则有If the sampling of f(x) at integer points and half-integer points is regarded as a discrete signal, that is, x(n)=f(n/2), then we have

C_0，n＝(x(2n)，x(2n+1))^T；C _{0, n} = (x(2n), x(2n+1)) ^T ;

其次：进行正交多小波分解，对向量化后的行和列采用张量积的形式进行两次连续的多小波变换，设

Secondly: Carry out orthogonal multi-wavelet decomposition, carry out two consecutive multi-wavelet transforms in the form of tensor product for the vectorized rows and columns, set

C表示一幅M×N的图像，其中c_i，j表示像素值，0≤i≤M-1，0≤j≤N-1，然后对图像的行和列进行多小波分解，其中：C represents an M×N image, where c _{i, j} represent pixel values, 0≤i≤M-1, 0≤j≤N-1, and then perform multi-wavelet decomposition on the rows and columns of the image, where:

行方向的多小波分解如下：The multiwavelet decomposition in the row direction is as follows:

将C的每一行，按

n＝0，.....，N/2-1，i＝0，1，....，M-1的方式组成向量信号，对C_i(n)进行多小波分解，得到：For each row of C, press

n=0,..., N/2-1, i=0, 1,..., the mode of M-1 forms vector signal, carries out multi-wavelet decomposition to C _i (n), obtains:

${C C}_{i i}^{L L} ((m m)) = = - - \frac{11}{\sqrt{22}} {P P}_{n no - - 22 m m} {C C}_{i i} ((n no)) = = [\begin{matrix} {c c}_{i i,, 22 m m}^{L L} \\ {c c}_{i i,, 22 m m + + 11}^{L L} \end{matrix}]$

${C C}_{i i}^{H h} ((m m)) = = - - \frac{11}{\sqrt{22}} {Q Q}_{n no - - 22 m m} {C C}_{i i} ((n no)) = = [\begin{matrix} {c c}_{i i,, 22 m m}^{H h} \\ {c c}_{i i,, 22 m m + + 11}^{H h} \end{matrix}]$

m＝0，1....，N/4-1，i＝0，.....，M-1；经过行方向的多小波分解，得到：m=0, 1..., N/4-1, i=0,..., M-1; after multi-wavelet decomposition in row direction, get:

${C C}_{row row} = = [\begin{matrix} {c c}_{0,0 0,0}^{L L} & . . . . . . & {c c}_{00,, \frac{N N}{22} - - 11}^{L L} & {c c}_{0,0 0,0}^{H h} & . . . . . . & {c c}_{00,, \frac{N N}{22} - - 11}^{H h} \\ {c c}_{1,0 1,0}^{L L} & . . . . . . & {c c}_{11,, \frac{N N}{22} - - 11}^{L L} & {c c}_{1,0 1,0}^{H h} & . . . . . . & {c c}_{11,, \frac{N N}{22} - - 11}^{H h} \\ . . & . . & . . & . . & . . & . . \\ . . & . . & . . & . . & . . & . . \\ . . & . . & . . & . . & . . & . . \\ {c c}_{M m - - 1,0 1,0}^{L L} & . . . . . . & {c c}_{M m - - 11,, \frac{N N}{22} - - 11}^{L L} & {c c}_{M m - - 1,0 1,0}^{H h} & . . . . . . & {c c}_{M m - - 11,, \frac{N N}{22} - - 11}^{H h} \end{matrix}]$

C_row表示图像C经过行方向的多小波分解后的小波系数；C _row represents the wavelet coefficients of the image C after multi-wavelet decomposition in the row direction;

列方向的多小波分解如下：The multiwavelet decomposition in the column direction is as follows:

将C_row的每一列，按如下的方式组成向量信号，Each column of C _row is composed of vector signals in the following way,

${C C}_{j j}^{L L} ((n no)) = = [\begin{matrix} {c c}_{22 n no,, j j}^{L L} \\ {c c}_{22 n no + + 11,, j j}^{L L} \end{matrix}],,$ ${C C}_{j j}^{H h} ((n no)) = = [\begin{matrix} {c c}_{22 n no,, j j}^{H h} \\ {c c}_{22 n no + + 11,, j j}^{H h} \end{matrix}],,$

n＝0，...，M/2-1，j＝0，1，...，N/2-1；然后分别对

进行多小波分解，得到：n=0,..., M/2-1, j=0, 1,..., N/2-1; then respectively

Perform multi-wavelet decomposition to get:

${C C}_{j j}^{LL LL} ((m m)) = = \frac{11}{\sqrt{22}} \underset{n no}{Σ Σ} {P P}_{n no - - 22 m m} {C C}_{j j}^{L L} ((n no)) = = [\begin{matrix} {c c}_{22 m m,, j j}^{LL LL} \\ {c c}_{22 m m + + 11,, j j}^{LL LL} \end{matrix}]$

${C C}_{j j}^{LH LH} ((m m)) = = \frac{11}{\sqrt{22}} \underset{n no}{Σ Σ} {Q Q}_{n no - - 22 m m} {C C}_{j j}^{L L} ((n no)) = = [\begin{matrix} {c c}_{22 m m,, j j}^{LH LH} \\ {c c}_{22 m m + + 11,, j j}^{LH LH} \end{matrix}]$

${C C}_{j j}^{HL HL} ((m m)) = = \frac{11}{\sqrt{22}} \underset{n no}{Σ Σ} {P P}_{n no - - 22 m m} {C C}_{j j}^{H h} ((n no)) = = [\begin{matrix} {c c}_{22 m m,, j j}^{HL HL} \\ {c c}_{22 m m + + 11,, j j}^{HL HL} \end{matrix}]$

${C C}_{j j}^{HH HH} ((m m)) = = \frac{11}{\sqrt{22}} \underset{n no}{Σ Σ} {Q Q}_{n no - - 22 m m} {C C}_{j j}^{H h} ((n no)) = = [\begin{matrix} {c c}_{22 m m,, j j}^{HH HH} \\ {c c}_{22 m m + + 11,, j j}^{HH HH} \end{matrix}]$

经过列方向的多小波分解即可得到图像C的一层小波分解，即：After multi-wavelet decomposition in the column direction, one layer of wavelet decomposition of image C can be obtained, namely:

${C C}_{col col} = =$

$[\begin{matrix} {c c}_{0,0 0,0}^{LL LL} & . . . . . . & {c c}_{00,, N N / / 22 - - 11}^{LL LL} & {c c}_{0,0 0,0}^{HL HL} & . . . . . . & {c c}_{00,, N N / / 22 - - 11}^{HL HL} \\ . . & . . & . . & . . & . . & . . \\ . . & . . & . . & . . & . . & . . \\ . . & . . & . . & . . & . . & . . \\ {c c}_{M m / / 22 - - 1,0 1,0}^{LL LL} & . . . . . . & {c c}_{M m / / 22 - - 11,, N N / / 22 - - 11}^{LL LL} & {c c}_{M m / / 22 - - 1,0 1,0}^{HL HL} & . . . . . . & {c c}_{M m / / 22 - - 11,, N N / / 22 - - 11}^{HL HL} \\ {c c}_{0,0 0,0}^{LH LH} & . . . . . . & {c c}_{00,, N N / / 22 - - 11}^{LH LH} & {c c}_{0,0 0,0}^{HH HH} & . . . . . . & {c c}_{00,, N N / / 22 - - 11}^{HH HH} \\ . . & . . & . . & . . & . . & . . \\ . . & . . & . . & . . & . . & . . \\ . . & . . & . . & . . & . . & . . \\ {c c}_{M m / / 22 - - 1,0 1,0}^{LH LH} & . . . . . . & {c c}_{M m / / 22 - - 11,, N N / / 22 - - 11}^{LH LH} & {c c}_{M m / / 22 - - 1,0 1,0}^{HH HH} & . . . . . . & {c c}_{M m / / 22 - - 11,, N N / / 22 - - 11}^{HH HH} \end{matrix}]$

$= = [\begin{matrix} LL LL & HL HL \\ LH LH & HH HH \end{matrix}] . .$

其中：C_col表示图像C经过一层小波分解之后的小波系数。Among them: C _col represents the wavelet coefficient of the image C after one layer of wavelet decomposition.

所述对进行小波分解后的小波系数进行插值处理包括：The interpolation processing of the wavelet coefficients after wavelet decomposition includes:

将三个高频带小波系数插入零，插入零后获取水平垂直方向各放大两倍的高频予频带系数，表格内括号内数字为像素点位置索引(x，y)；Insert zero into the three high-frequency band wavelet coefficients, and obtain the high-frequency pre-band coefficients that are enlarged twice in the horizontal and vertical directions after inserting zeros. The numbers in brackets in the table are the pixel position indexes (x, y);

在原图像乘上一个系数n作为低频带小波系数，在这里令n＝2。Multiply a coefficient n on the original image as the wavelet coefficient of the low frequency band, and let n=2 here.

所述将进行插值处理后的正交多小波重建得到缩放的图像包括：The image zoomed by the orthogonal multi-wavelet reconstruction after interpolation processing includes:

进行列方向的重建如下：Perform column-wise reconstruction as follows:

采用快速正交小波的重建方法可得到

与

即：Using fast orthogonal wavelet reconstruction method can get

and

Right now:

${C C}_{j j}^{L L} ((n no)) = = \frac{11}{\sqrt{22}} \underset{m m}{Σ Σ} {P P}_{n no - - 22 m m}^{T T} {C C}_{j j}^{LL LL} ((m m)) + + \frac{11}{\sqrt{22}} \underset{m m}{Σ Σ} {Q Q}_{n no - - 22 m m}^{T T} {C C}_{j j}^{LH LH} ((m m))$

${C C}_{j j}^{H h} ((n no)) = = \frac{11}{\sqrt{22}} \underset{m m}{Σ Σ} {P P}_{n no - - 22 m m}^{T T} {C C}_{j j}^{HL HL} ((m m)) + + \frac{11}{\sqrt{22}} \underset{m m}{Σ Σ} {Q Q}_{n no - - 22 m m}^{T T} {C C}_{j j}^{HH HH} ((m m))$

n＝0，1，...，M/2-1，j＝0，1，....，N/2-1.由

与

形成中间重建图像C_row。n=0, 1,..., M/2-1, j=0, 1,..., N/2-1. By

and

An intermediate reconstructed image C _row is formed.

进行行方向的重建如下：Perform row-wise reconstruction as follows:

采用快速正交小波的重建方法可得到C_i(n)，即：Using fast orthogonal wavelet reconstruction method can get C _i (n), namely:

${C C}_{i i} ((n no)) = = \frac{11}{\sqrt{22}} \underset{m m}{Σ Σ} {P P}_{n no - - 22 m m}^{T T} {C C}_{i i}^{L L} ((m m)) + + \frac{11}{\sqrt{22}} \underset{m m}{Σ Σ} {Q Q}_{n no - - 22 m m}^{T T} {C C}_{i i}^{H h} ((m m))$

n＝0，1，...，N/2-1，i＝0，...，M-1；最后由C_i(n)可以重建的得到缩放的图像C。n=0 _, 1, . . . , N/2-1, i=0, .

所述缩放的图像C中进行图像的缩小包括：Reducing the image in the zoomed image C includes:

通过多小波分解，将图像分解成各个分辨率上的低频成分以及一系列的高频成分，通过适当的方法求得低频成分，得到源图像的一个缩小2倍的近似图像，以此类推，通过K次分解得到原图像的一个缩小2^k倍的近似图像。Through multi-wavelet decomposition, the image is decomposed into low-frequency components and a series of high-frequency components at various resolutions, and the low-frequency components are obtained by appropriate methods to obtain an approximate image that is reduced by 2 times, and so on, through K decompositions get an approximate image that is 2 ^k times smaller than the original image.

所述缩放的图像C中进行图像的放大包括：Enlarging the image in the zoomed image C includes:

利用中的系数进行处理得到近似的高频成分{LH，HL，HH}，然后按照重建的方法得到放大了2倍的图像，依此类推，通过k次重建得到原图像的一个放大2^k倍的近似图像。use The coefficients in are processed to obtain approximate high-frequency components {LH, HL, HH}, and then a 2-fold enlarged image is obtained according to the reconstruction method, and so on, and a 2 ^k- fold enlarged image of the original image is obtained through k reconstructions approximate image.

上述技术方案可以看出，由于本发明实施例采用基于插值正交多小波的视频图像缩放方法，因此能够增强图像的边缘，克服了缩放图像边缘模糊，从而提高了图像的对比度、清晰度。From the above technical solution, it can be seen that since the embodiment of the present invention adopts the video image scaling method based on interpolation orthogonal multi-wavelet, the edge of the image can be enhanced, the edge blur of the scaled image can be overcome, and the contrast and definition of the image can be improved.

附图说明Description of drawings

为了更清楚地说明本发明实施例或现有技术中的技术方案，下面将对实施例或现有技术描述中所需要使用的附图作简单地介绍，显而易见地，下面描述中的附图仅仅是本发明的一些实施例，对于本领域普通技术人员来讲，在不付出创造性劳动性的前提下，还可以根据这些附图获得其他的附图。In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings that need to be used in the description of the embodiments or the prior art. Obviously, the accompanying drawings in the following description are only These are some embodiments of the present invention. For those skilled in the art, other drawings can also be obtained according to these drawings without any creative effort.

图1为本发明实施例中的基于插值正交多小波的视频图像缩放的方法流程图；Fig. 1 is the method flowchart of the video image scaling based on interpolation orthogonal multi-wavelet in the embodiment of the present invention;

图2为本发明实施例中的基于插值正交多小波的视频图像缩放系统架构图；Fig. 2 is the frame diagram of the video image scaling system based on interpolation orthogonal multi-wavelet in the embodiment of the present invention;

图3为本发明实施例中的正交多小波分解流程图；Fig. 3 is the flow chart of orthogonal multi-wavelet decomposition in the embodiment of the present invention;

图4为本发明实施例中的插零过程示意图。Fig. 4 is a schematic diagram of a zero insertion process in an embodiment of the present invention.

具体实施方式Detailed ways

下面将结合本发明实施例中的附图，对本发明实施例中的技术方案进行清楚、完整地描述，显然，所描述的实施例仅仅是本发明一部分实施例，而不是全部的实施例。基于本发明中的实施例，本领域普通技术人员在没有作出创造性劳动前提下所获得的所有其它实施例，都属于本发明保护的范围。The following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only some, not all, embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by persons of ordinary skill in the art without creative efforts fall within the protection scope of the present invention.

本发明提供一种基于插值正交多小波的视频图像缩放方法，小波变换是在傅立叶变换基础之上发展起来的，现在已经广泛应用于图像处理、语音和视频处理以及数字信号处理等领域。在工程实际领域中，由于小波变换具有良好的特性，在众多实际应用中都能得到很好的应用。利用小波变换的放大、缩小和平移的数学显微镜功能，可以方便的产生各种分辨率的图像。多小波是指由两个或两个以上函数作为尺度函数生成的小波。已有的成果表明多小波可同时满足对称性、紧支撑性、高阶消失矩和正交性，所以在信号处理等应用方面比单小波更有优势。该方法能够增强图像的边缘，克服了分形插值引起的图像边缘模糊，从而提高了图像的对比度、清晰度。放大后图像轮廓光滑清晰，图像内部的对比度有所增强。The invention provides a video image scaling method based on interpolation orthogonal multi-wavelet. The wavelet transform is developed on the basis of Fourier transform and has been widely used in the fields of image processing, voice and video processing, and digital signal processing. In the field of engineering practice, due to the good characteristics of wavelet transform, it can be well applied in many practical applications. Images of various resolutions can be easily produced by using the mathematical microscope function of zooming in, zooming out and translation of wavelet transform. Multiwavelet refers to wavelets generated by two or more functions as scaling functions. Existing results show that multi-wavelet can satisfy symmetry, tight support, high-order vanishing moments and orthogonality at the same time, so it has more advantages than single wavelet in signal processing and other applications. This method can enhance the edge of the image, overcome the blurring of the edge of the image caused by fractal interpolation, and thus improve the contrast and clarity of the image. After zooming in, the outline of the image is smooth and clear, and the contrast inside the image is enhanced.

图1示出了本发明，实施例中的基于插值正交多小波的视频图像缩放的方法流程图包括：Fig. 1 shows the present invention, the method flowchart of the video image scaling based on interpolation orthogonal multi-wavelet in the embodiment comprises:

S101：获取原始图像数据；S101: Acquiring original image data;

S102：对原始图像数据进行正交多小波分解；S102: Perform orthogonal multi-wavelet decomposition on the original image data;

S103：对进行小波分解后的小波系数进行插值处理；S103: Perform interpolation processing on the wavelet coefficients after wavelet decomposition;

S104：将进行插值处理后的正交多小波重建得到缩放的图像；S104: Reconstruct the interpolated orthogonal multi-wavelet to obtain a scaled image;

S105：将缩放后的图像输出。S105: Outputting the scaled image.

本发明实施例提供一种基于插值正交多小波的视频图像缩放的方法，能够得到较好视觉效果的图像缩放，即缩小了的图像能较好地保持原图像的结构特征，与原图像有一致的边缘；放大了的图像也具有较高的清晰度，克服了用内插方法实现图像缩放在这两方面的缺陷。本发明通过将图像数据向量化之后进行正交多小波分解，然后通过对分解后的小波系数的进行处理，最后将分解经过处理后的数据进行正交多小波重建达到将原图像放大或缩小的目的。以下分别进行详细说明。Embodiments of the present invention provide a video image scaling method based on interpolation orthogonal multi-wavelets, which can obtain image scaling with better visual effects, that is, the reduced image can better maintain the structural characteristics of the original image, and is similar to the original image. Consistent edges; the magnified image also has high definition, which overcomes the defects in the two aspects of image scaling realized by the interpolation method. The present invention performs orthogonal multi-wavelet decomposition after vectorizing the image data, then processes the decomposed wavelet coefficients, and finally performs orthogonal multi-wavelet reconstruction on the decomposed and processed data to enlarge or reduce the original image. Purpose. Each will be described in detail below.

图2示出了本发明实施例中的基于插值正交多小波的视频图像缩放系统架构图，类似于计算机中内存与寄存器的两级存储结构，图像输入缓存也使用两级缓存的结构：第一级容量较大，存储图像的区域像素数据；第二级容量较小，只存储参与计算的像素数据。第一级缓存器为双口RAM，每个双口RAM的容量为一行目标分辨率的图像数据大小，所述第二缓存器为双口RAM。双口RAM的容量为五行目标分辨率的图像数据大小。第一级缓存器与第二缓存器实现可靠和稳定的跨时针的数据传输和数据处理同步。由于定标器输入的图像分辨率可能有很多种，因此对应的输入像素时针也不是固定的，而无论是进行图像放大还是缩小，缩放结果的图像数据都与输入图像分辨率不同，因此显示时针和输入像素时针肯定是不同的，这就意味着需要数据缓冲器来进行跨时针的数据传输。系统从第一级缓存器取得原始图像数据，然后将经过缩放模块的图像数据写入第二级缓冲器。Fig. 2 shows the frame diagram of the video image scaling system based on interpolation orthogonal multi-wavelet in the embodiment of the present invention, similar to the two-level storage structure of memory and register in the computer, the image input cache also uses the structure of two-level cache: the first The first level has a large capacity and stores the pixel data of the image area; the second level has a small capacity and only stores the pixel data involved in the calculation. The first-level buffer is a dual-port RAM, and the capacity of each dual-port RAM is the size of a line of image data with a target resolution. The second buffer is a dual-port RAM. The capacity of the dual-port RAM is the image data size of the five-line target resolution. The first-level buffer and the second buffer realize reliable and stable cross-clockwise data transmission and data processing synchronization. Since there may be many kinds of image resolutions input by the scaler, the corresponding input pixel hour hand is not fixed, and whether the image is enlarged or reduced, the image data of the scaling result is different from the input image resolution, so the display hour hand It is definitely different from the input pixel clock, which means that a data buffer is required for data transfer across the clock. The system obtains the original image data from the first-level buffer, and then writes the image data passed through the scaling module into the second-level buffer.

图像缩放模块image scaling module

如图2所示，缩放模块包括正交多小波分解，小波系数的处理和正交多小波重建三个步骤。具体如下：As shown in Figure 2, the scaling module includes three steps of orthogonal multi-wavelet decomposition, processing of wavelet coefficients and reconstruction of orthogonal multi-wavelets. details as follows:

因为多小波变换只适用于向量信号，因此，要对图像进行多小波变换，必须首先对图像的行和列进行向量化。Because the multiwavelet transform is only applicable to vector signals, to perform multiwavelet transform on an image, the rows and columns of the image must be vectorized first.

一维离散信号的向量化如下：The vectorization of a one-dimensional discrete signal is as follows:

其中

in

C_0，n＝(x(2n)，x(2n+1))^T C _{0, n} = (x(2n), x(2n+1)) ^T

正交多小波分解Orthogonal Multiwavelet Decomposition

详细可以参阅图2中的正交多小波分解流程图，包括如下：For details, please refer to the flow chart of orthogonal multi-wavelet decomposition in Figure 2, including the following:

A、进行行向量化；A. Carry out row vectorization;

B、进行行方向正交多小波分解；B. Carry out row direction orthogonal multi-wavelet decomposition;

C、进行行分解系数向量化；C. Perform row decomposition coefficient vectorization;

D、列方向正交多小波分解；D. Orthogonal multi-wavelet decomposition in the column direction;

E、一层小波分解；E. One-layer wavelet decomposition;

F、是否满足缩放要求，满足则进行G，否则进行H；F. Whether the scaling requirements are met, if so, proceed to G, otherwise proceed to H;

G、输出分解系数；G. Output decomposition coefficient;

H、提取低频系数。H. Extract low-frequency coefficients.

首先，对向量化后的行和列一般采用张量积的形式进行两次连续的多小波变换。设

First of all, two consecutive multiwavelet transforms are generally performed on the vectorized rows and columns in the form of tensor products. set up

C表示一幅M×N的图像，其中c_i，j表示像素值，0≤i≤M-1，0≤j≤N-1，然后对图像的行和列按以下方法进行多小波分解。C represents an M×N image, where c _{i and j} represent pixel values, 0≤i≤M-1, 0≤j≤N-1, and then perform multi-wavelet decomposition on the rows and columns of the image as follows.

行方向的多小波分解Multiwavelet Decomposition in Row Direction

将C的每一行，按n＝0，.....，N/2-1，i＝0，1，....，M-1的方式组成向量信号，对C_i(n)进行多小波分解，得到：For each row of C, press n=0,..., N/2-1, i=0, 1,..., the mode of M-1 forms vector signal, carries out multi-wavelet decomposition to C _i (n), obtains:

m＝0，1....，N/4-1，i＝0，.....，M-1。于是，经过行方向的多小波分解，得到：m=0, 1..., N/4-1, i=0,..., M-1. Then, after multi-wavelet decomposition in the row direction, we get:

C_row表示图像C经过行方向的多小波分解后的小波系数。C _row represents the wavelet coefficients of the image C after multi-wavelet decomposition in the row direction.

列方向的多小波分解Multiwavelet Decomposition in Column Direction

${C C}_{j j}^{H h} ((n no)) = = [\begin{matrix} {c c}_{22 n no,, j j}^{H h} \\ {c c}_{22 n no + + 11,, j j}^{H h} \end{matrix}],,$ ${C C}_{j j}^{H h} ((n no)) = = [\begin{matrix} {c c}_{22 n no,, j j}^{H h} \\ {c c}_{22 n no + + 11,, j j}^{H h} \end{matrix}],,$

n＝0，...，M/2-1，j＝0，1，...，N/2-1。然后分别对

进行多小波分解，得到：n=0, . . . , M/2-1, j=0, 1, . . . , N/2-1. Then respectively for

Perform multi-wavelet decomposition to get:

最后，经过列方向的多小波分解即可得到图像C的一层小波分解，即：Finally, after multi-wavelet decomposition in the column direction, one layer of wavelet decomposition of image C can be obtained, namely:

${C C}_{col col} = =$

$= = [\begin{matrix} LL LL & HL HL \\ LH LH & HH HH \end{matrix}] . .$

C_col表示图像C经过一层小波分解之后的小波系数。C _col represents the wavelet coefficient of the image C after one layer of wavelet decomposition.

实际应用中如果要将图像缩放成2^k倍，则需要将图像进行k层多正交小波分解即可得到，只需要对每次变换后的LL子图像再次进行多小波分解即可。In practical applications, if the image is to be scaled to 2 ^k times, it is necessary to decompose the image with k-level multi-orthogonal wavelets, and it is only necessary to perform multi-wavelet decomposition on each transformed LL sub-image.

小波系数的处理Processing of wavelet coefficients

第一步：将三个高频带小波系数按图4所示的方式插入零，插入零后获取水平垂直方向各放大两倍的高频予频带系数，表格内括号内数字为像素点位置索引(x，y)；Step 1: Insert zeros into the three high-frequency band wavelet coefficients as shown in Figure 4. After inserting zeros, obtain the high-frequency pre-band coefficients that are enlarged twice in the horizontal and vertical directions. The numbers in brackets in the table are the pixel position indexes (x ,y);

第二步：为防止能量损失，使用原图像乘上一个系数n作为低频带小波系数，在这里令n＝2；The second step: in order to prevent energy loss, use the original image multiplied by a coefficient n as the low-frequency band wavelet coefficient, let n=2 here;

正交多小波重建Orthogonal Multiwavelet Reconstruction

将经过插值正交小波分解的图像进行插值正交多小波重建得到缩放的图像。具体步骤如下：The image decomposed by interpolation orthogonal wavelet is reconstructed by interpolation orthogonal multi-wavelet to obtain the scaled image. Specific steps are as follows:

(1)进行列方向的重建(1) Carry out reconstruction in the column direction

采用快速正交小波的重建方法可得到

与

即：Using fast orthogonal wavelet reconstruction method can get

and

Right now:

n＝0，1，...，M/2-1，j＝0，1，....，N/2-1。由

与

形成中间重建图像C_row。n=0, 1, . . . , M/2-1, j=0, 1, . . . , N/2-1. Depend on

and

An intermediate reconstructed image C _row is formed.

(2)行方向的重建(2) Reconstruction in row direction

n＝0，1，...，N/2-1，i＝0，...，M-1。最后由C_i(n)可以重建的得到缩放的图像C。n=0, 1, . . . , N/2-1, i=0, . . . , M-1. Finally, the scaled image C can be reconstructed from C _i (n).

以上过程的图像放大和缩小的相关操作：Related operations of image zoom-in and zoom-out in the above process:

图像的缩小image downscaling

通过多小波分解，将图像分解成各个分辨率上的平滑版本(低频成分)以及一系列的细节版本(高频成分)，通过适当的方法求得低频成分，可得到源图像的一个缩小2倍的近似图像，以此类推，通过K次分解可得到原图像的一个缩小2^k倍的近似图像。Through multi-wavelet decomposition, the image is decomposed into a smooth version (low frequency component) at each resolution and a series of detailed versions (high frequency component), and the low frequency component is obtained by an appropriate method, and a 2 times reduction of the source image can be obtained The approximate image of , and so on, can obtain an approximate image reduced by 2 ^k times of the original image through K decomposition.

图像的放大image enlargement

放大过程的关键是如何求得适当的高频成分{LH，HL，HH}，这里利用

中的系数进行处理得到近似的高频成分{LH，HL，HH}，然后按照重建的方法得到放大了2倍的图像，依此类推，通过k次重建可得到原图像的一个放大2^k倍的近似图像。The key to the amplification process is how to obtain the appropriate high-frequency components {LH, HL, HH}, here use

The coefficients in are processed to obtain approximate high-frequency components {LH, HL, HH}, and then an image enlarged by 2 times is obtained according to the reconstruction method, and so on, and a magnification of the original image by 2 ^k times can be obtained through k reconstructions Approximate image of .

综上，通过实施本发明实施例，由于本发明实施例采用基于插值正交多小波的视频图像缩放方法，因此能够增强图像的边缘，克服了缩放图像边缘模糊，从而提高了图像的对比度、清晰度。In summary, through the implementation of the embodiment of the present invention, since the embodiment of the present invention adopts the video image scaling method based on interpolation orthogonal multi-wavelet, the edge of the image can be enhanced, and the blurred edge of the scaled image can be overcome, thereby improving the contrast and clarity of the image. Spend.

以上对本发明实施例所提供的基于插值正交多小波的视频图像缩放的方法进行了详细介绍，本文中应用了具体个例对本发明的原理及实施方式进行了阐述，以上实施例的说明只是用于帮助理解本发明的方法及其核心思想；同时，对于本领域的一般技术人员，依据本发明的思想，在具体实施方式及应用范围上均会有改变之处，综上所述，本说明书内容不应理解为对本发明的限制。The method for zooming video images based on interpolation orthogonal multi-wavelets provided by the embodiments of the present invention has been described in detail above. The principles and implementation methods of the present invention have been described by using specific examples in this paper. The descriptions of the above embodiments are only used To help understand the method of the present invention and its core idea; at the same time, for those of ordinary skill in the art, according to the idea of the present invention, there will be changes in the specific implementation and scope of application. In summary, this specification The content should not be construed as a limitation of the invention.

Claims

1. A method for video image zooming based on interpolation orthogonal multi-wavelet, is characterized in that, comprises:

Get raw image data;

Orthogonal multi-wavelet decomposition is performed on the original image data;

Perform interpolation processing on the wavelet coefficients after wavelet decomposition;

Reconstruct the interpolated orthogonal multi-wavelet to obtain a scaled image;

Output the scaled image.

2. The method according to claim 1, wherein said carrying out orthogonal multi-wavelet decomposition to original image data comprises:

First: vectorize the rows and columns of the image, where: the vectorization of a one-dimensional discrete signal is as follows:

&ForAll; f (t) &Element; V_{0},

then there is

According to the interpolation properties of multi-scale function and multi-wavelet:

C _0,n =(c _0,0,n ,c _1,0,n ) ^T =(f(n),f(n+1/2)) ^T

in

If the sampling of f(x) at integer points and half-integer points is regarded as a discrete signal, that is, x(n)=f(n/2), then we have

C _{0, n} = (x(2n), x(2n+1)) ^T ;

C represents an M×N image, where c _{i, j} represent pixel values, 0≤i≤M-1, 0≤j≤N-1, and then perform multi-wavelet decomposition on the rows and columns of the image, where:

The multiwavelet decomposition in the row direction is as follows:

For each row of C, press

{C C}_{i i}^{L L} ((m m)) = = - - \frac{11}{\sqrt{22}} {P P}_{n no - - 22 m m} {C C}_{i i} ((n no)) = = [\begin{matrix} {c c}_{i i,, 22 m m}^{L L} \\ {c c}_{i i,, 22 m m + + 11}^{L L} \end{matrix}]

{C C}_{i i}^{H h} ((m m)) = = - - \frac{11}{\sqrt{22}} {Q Q}_{n no - - 22 m m} {C C}_{i i} ((n no)) = = [\begin{matrix} {c c}_{i i,, 22 m m}^{H h} \\ {c c}_{i i,, 22 m m + + 11}^{H h} \end{matrix}]

m=0, 1..., N/4-1, i=0,..., M-1; after multi-wavelet decomposition in row direction, get:

{C C}_{row row} = = [\begin{matrix} {c c}_{0,0 0,0}^{L L} & . . . . . . & {c c}_{00,, \frac{N N}{22} - - 11}^{L L} & {c c}_{0,0 0,0}^{H h} & . . . . . . & {c c}_{00,, \frac{N N}{22} - - 11}^{H h} \\ {c c}_{1,0 1,0}^{L L} & . . . . . . & {c c}_{11,, \frac{N N}{22} - - 11}^{L L} & {c c}_{1,0 1,0}^{H h} & . . . . . . & {c c}_{11,, \frac{N N}{22} - - 11}^{H h} \\ . . & . . & . . & . . & . . & . . \\ . . & . . & . . & . . & . . & . . \\ . . & . . & . . & . . & . . & . . \\ {c c}_{M m - - 1,0 1,0}^{L L} & . . . . . . & {c c}_{M m - - 11,, \frac{N N}{22} - - 11}^{L L} & {c c}_{M m - - 1,0 1,0}^{H h} & . . . . . . & {c c}_{M m - - 11,, \frac{N N}{22} - - 11}^{H h} \end{matrix}]

C _row represents the wavelet coefficients of the image C after multi-wavelet decomposition in the row direction;

The multiwavelet decomposition in the column direction is as follows:

Each column of C _row is composed of vector signals in the following way,

{C C}_{j j}^{L L} ((n no)) = = [\begin{matrix} {c c}_{22 n no,, j j}^{L L} \\ {c c}_{22 n no + + 11,, j j}^{L L} \end{matrix}],,

{C C}_{j j}^{H h} ((n no)) = = [\begin{matrix} {c c}_{22 n no,, j j}^{H h} \\ {c c}_{22 n no + + 11,, j j}^{H h} \end{matrix}],,

n=0,..., M/2-1, j=0, 1,..., N/2-1; then respectively

Perform multi-wavelet decomposition to get:

{C C}_{j j}^{LL LL} ((m m)) = = \frac{11}{\sqrt{22}} \underset{n no}{Σ Σ} {P P}_{n no - - 22 m m} {C C}_{j j}^{L L} ((n no)) = = [\begin{matrix} {c c}_{22 m m,, j j}^{LL LL} \\ {c c}_{22 m m + + 11,, j j}^{LL LL} \end{matrix}]

{C C}_{j j}^{LH LH} ((m m)) = = \frac{11}{\sqrt{22}} \underset{n no}{Σ Σ} {Q Q}_{n no - - 22 m m} {C C}_{j j}^{L L} ((n no)) = = [\begin{matrix} {c c}_{22 m m,, j j}^{LH LH} \\ {c c}_{22 m m + + 11,, j j}^{LH LH} \end{matrix}]

{C C}_{j j}^{HL HL} ((m m)) = = \frac{11}{\sqrt{22}} \underset{n no}{Σ Σ} {P P}_{n no - - 22 m m} {C C}_{j j}^{H h} ((n no)) = = [\begin{matrix} {c c}_{22 m m,, j j}^{HL HL} \\ {c c}_{22 m m + + 11,, j j}^{HL HL} \end{matrix}]

{C C}_{j j}^{HH HH} ((m m)) = = \frac{11}{\sqrt{22}} \underset{n no}{Σ Σ} {Q Q}_{n no - - 22 m m} {C C}_{j j}^{H h} ((n no)) = = [\begin{matrix} {c c}_{22 m m,, j j}^{HH HH} \\ {c c}_{22 m m + + 11,, j j}^{HH HH} \end{matrix}]

After multi-wavelet decomposition in the column direction, one layer of wavelet decomposition of image C can be obtained, namely:

{C C}_{col col} = =

[\begin{matrix} {c c}_{0,0 0,0}^{LL LL} & . . . . . . & {c c}_{00,, N N / / 22 - - 11}^{LL LL} & {c c}_{0,0 0,0}^{HL HL} & . . . . . . & {c c}_{00,, N N / / 22 - - 11}^{HL HL} \\ . . & . . & . . & . . & . . & . . \\ . . & . . & . . & . . & . . & . . \\ . . & . . & . . & . . & . . & . . \\ {c c}_{M m / / 22 - - 1,0 1,0}^{LL LL} & . . . . . . & {c c}_{M m / / 22 - - 11,, N N / / 22 - - 11}^{LL LL} & {c c}_{M m / / 22 - - 1,0 1,0}^{HL HL} & . . . . . . & {c c}_{M m / / 22 - - 11,, N N / / 22 - - 11}^{HL HL} \\ {c c}_{0,0 0,0}^{LH LH} & . . . . . . & {c c}_{00,, N N / / 22 - - 11}^{LH LH} & {c c}_{0,0 0,0}^{HH HH} & . . . . . . & {c c}_{00,, N N / / 22 - - 11}^{HH HH} \\ . . & . . & . . & . . & . . & . . \\ . . & . . & . . & . . & . . & . . \\ . . & . . & . . & . . & . . & . . \\ {c c}_{M m / / 22 - - 1,0 1,0}^{LH LH} & . . . . . . & {c c}_{M m / / 22 - - 11,, N N / / 22 - - 11}^{LH LH} & {c c}_{M m / / 22 - - 1,0 1,0}^{HH HH} & . . . . . . & {c c}_{M m / / 22 - - 11,, N N / / 22 - - 11}^{HH HH} \end{matrix}]

= = [\begin{matrix} LL LL & HL HL \\ LH LH & HH HH \end{matrix}] . .

Among them: C _col represents the wavelet coefficient of the image C after one layer of wavelet decomposition.

3. The method according to claim 2, wherein said carrying out interpolation processing to the wavelet coefficients after wavelet decomposition comprises:

Insert zero into the three high-frequency band wavelet coefficients, and obtain the high-frequency pre-band coefficients that are enlarged twice in the horizontal and vertical directions after inserting zeros. The numbers in brackets in the table are the pixel position indexes (x, y);

Multiply a coefficient n on the original image as the wavelet coefficient of the low frequency band, where n=2.

4. method as claimed in claim 3, is characterized in that, described will carry out the image that the orthogonal multi-wavelet reconstruction after interpolation process obtains scaling comprises:

Perform column-wise reconstruction as follows:

Using fast orthogonal wavelet reconstruction method can get

and

Right now:

{C C}_{j j}^{L L} ((n no)) = = \frac{11}{\sqrt{22}} \underset{m m}{Σ Σ} {P P}_{n no - - 22 m m}^{T T} {C C}_{j j}^{LL LL} ((m m)) + + \frac{11}{\sqrt{22}} \underset{m m}{Σ Σ} {Q Q}_{n no - - 22 m m}^{T T} {C C}_{j j}^{LH LH} ((m m))

{C C}_{j j}^{H h} ((n no)) = = \frac{11}{\sqrt{22}} \underset{m m}{Σ Σ} {P P}_{n no - - 22 m m}^{T T} {C C}_{j j}^{HL HL} ((m m)) + + \frac{11}{\sqrt{22}} \underset{m m}{Σ Σ} {Q Q}_{n no - - 22 m m}^{T T} {C C}_{j j}^{HH HH} ((m m))

n=0, 1,..., M/2-1, j=0, 1,..., N/2-1. By

and

An intermediate reconstructed image C _row is formed.

Perform row-wise reconstruction as follows:

Using fast orthogonal wavelet reconstruction method can get C _i (n), namely:

{C C}_{i i} ((n no)) = = \frac{11}{\sqrt{22}} \underset{m m}{Σ Σ} {P P}_{n no - - 22 m m}^{T T} {C C}_{i i}^{L L} ((m m)) + + \frac{11}{\sqrt{22}} \underset{m m}{Σ Σ} {Q Q}_{n no - - 22 m m}^{T T} {C C}_{i i}^{H h} ((m m))

n=0 _, 1, . . . , N/2-1, i=0, .

5. The method according to claim 4, characterized in that, shrinking the image in the zoomed image C comprises:

Through multi-wavelet decomposition, the image is decomposed into low-frequency components and a series of high-frequency components at various resolutions, and the low-frequency components are obtained by appropriate methods to obtain an approximate image that is reduced by 2 times, and so on, through K decompositions get an approximate image that is 2 ^k times smaller than the original image.

6. The method according to claim 4, characterized in that, image scaling is carried out in the zoomed image C

Magnification includes:

use

The coefficients in are processed to obtain approximate high-frequency components {LH, HL, HH}, and then a 2-fold enlarged image is obtained according to the reconstruction method, and so on, and a 2 ^k- fold enlarged image of the original image is obtained through k reconstructions approximate image.