CN101504775B - Roaming video automatic generation method based on image set - Google Patents

Roaming video automatic generation method based on image set Download PDF

Info

Publication number
CN101504775B
CN101504775B CN 200910096855 CN200910096855A CN101504775B CN 101504775 B CN101504775 B CN 101504775B CN 200910096855 CN200910096855 CN 200910096855 CN 200910096855 A CN200910096855 A CN 200910096855A CN 101504775 B CN101504775 B CN 101504775B
Authority
CN
Grant status
Grant
Patent type
Prior art keywords
image
embedded
difference
video
set
Prior art date
Application number
CN 200910096855
Other languages
Chinese (zh)
Other versions
CN101504775A (en )
Inventor
丛林
童若峰
Original Assignee
浙江大学
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Grant date

Links

Abstract

The invention provides a method for automatically generating roaming videos based on image sets, which comprises the following steps: firstly, local matching; secondly, global optimization; thirdly, image synthesis; and fourthly, connection of all the frames into a video for output. The method takes a series of images as input, finds out an optimum browsing sequence, and forms a continuously amplified image browsing video with visual continuity. The invention simultaneously provides a more vivid image browsing mode, which makes users browse images in a more vivid mode just like roaming in image scenes; as for a series of images of a tourist attraction, the method can generate a video for well connecting different scenes of the tourist attraction in series and take the video as a propaganda film of the tourist attraction; and the production method has high speed, does not require scene design and image selection with superabundant manpower and material resources, and has high automation degree in the whole video generation process.

Description

一种基于图像集的漫游视频自动生成方法 Roaming video image based method of automatically generating a set of

技术领域 FIELD

[0001] 本发明涉及一种基于放大形式的图像浏览方法。 [0001] The present invention relates to an image browsing method based on an enlarged form. 背景技术 Background technique

[0002] 基于放大形式的图像浏览方法隶属于非真实感图像视频处理领域,目前网上流行一种被成为zoomquilt的非真实感艺术表现形式:艺术家手绘一系图像,形成一个放大播放图像的flash视频,播放的视频在视觉上是连续的。 [0002] Non-photorealistic image belongs to the field of image viewing video processing method based on an enlarged form, currently known as a popular online forms of non-photorealistic artistic zoomquilt: the artist painted a series of images to form a magnified image of the flash video player , playing video is continuously visually. 但是要制作这样一个flash,需要艺术家精心设计并绘制场景,需要消耗大量人力。 But to make such a flash, the artist needs careful design and draw the scene, you need to consume a lot of manpower.

发明内容 SUMMARY

[0003] 本发明所要解决的技术问题是提供一种基于图像集的漫游视频自动生成方法,它能够基于计算机处理自动生成图像浏览视频。 [0003] The present invention solves the technical problem is to provide a method for automatically generating a video image based on the set of roaming, it is possible to automatically generate an image-based video browsing computer processing. 为此本发明采用以下技术方案:该方法包括以下步骤: To this end the present invention, the following technical solutions: the method comprising the steps of:

[0004] (1)本地匹配:设图像集中任意一幅图像为图像A,任意另一幅图像为图像B,将图像B缩小后,与图像A中的不同位置图像进行比较,计算出它们之间的图像差别,寻找所述图像A中与所述缩小的图像B之间图像差别最小的位置,将该位置作为所述缩小的图像B 在图像A中的嵌入位置,该最小图像差别作为两图像最终图像差别存入图像差别表;重复上述步骤,对图像集中的图像进行两两匹配,将获得的最小图像差别存入图像差别表; [0004] (1) Local Match: set any image set as an image of image A, the image of any other image B, after the reduced image B, and image A in different positions compared to calculate the sum thereof the difference between the image, to find the minimum difference between the position of the image of the image a in the image B is reduced, as the position of the embedded position B in the reduced image of image a, the difference image as a minimum two final image the image difference table is stored in the difference image; repeating the above steps, the image of the image set pairwise matching, the minimum difference image obtained is stored in the image difference table;

[0005] (2)全局优化:系统根据本地匹配得到的图像差别表,找出图像差别总和最小的原始图像排列序列,作为播放序列,其中,嵌入的图像排在被嵌入的图像的后面; [0005] (2) Global Optimization: The image difference table of the local system of matching obtained, the sum of the difference image to identify the minimum sequence of the original image arrangement, as the playback sequence, wherein the embedded image is embedded at the back of the image;

[0006] (3)图像合成:按照得到的播放序列,按前后次序,设序列中原始图像为i、i+1、 i+2……i+(nl)、i+n,将i+1缩小并按照本地匹配中获得的嵌入位置嵌入到i中,放大i和i+Ι的合成图像,直至i+Ι充满视窗;重复上述过程,进行i+Ι与i+2直至i+(nl)与i+n的处理;每个放大阶段的合成图像都作为前后原始图像之间的插值图像,所有图像输出视频的帧; [0006] (3) Image Synthesis: according to the play sequence obtained by the order before and after, provided the sequence of the original image is a i, i + 1, i + 2 ...... i + (nl), i + n, a i + 1 refine and embedded according to embed local matching obtained in a position to i in the enlarged combined image i and i + Ι until i + Ι filled windows; repeat the process, for i + Ι and i + 2 until i + (nl) and i + n processing; synthetic image are each amplification stage as the interpolation image between before and after the original image, all the images of the output video frame;

[0007] 最后,将所有帧连接成视频输出。 [0007] Finally, all frames are connected to the video output.

[0008] 基于放大形式的图像浏览方法,定义的图像差别包含以下要素:颜色差别,纹理差别,图像复杂性。 [0008] The image browsing method in an enlarged form based on the definition of the difference image comprises the following elements: color differences, texture difference image complexity.

[0009] 颜色是图像最显著的特征,与其他特征相比,颜色特征计算简单、性质稳定,对于旋转、平移、尺度变化都不敏感,表现出很强的鲁棒性。 [0009] The color image is the most significant feature, in comparison with other features, color feature calculation is simple, stable, for rotation, translation and scale change is not sensitive, exhibits robustness. 颜色特征包括颜色直方图、主要颜色、平均亮度等。 Color features include color histogram, the main color, the average luminance and the like. 本方法使用HSV颜色空间上的平均像素差平方(sum squared of difference)来衡量颜色差别 This method uses the average pixel difference square (sum squared of difference) in the HSV color space to measure the difference in color

[0010] 纹理分析一直是计算机视觉的一个重要研究方向,其方法可以大致分为统计方法和结构方法。 [0010] Texture analysis has been an important research direction of computer vision, its methods can be roughly divided into statistical methods and structural methods. 统计方法是对图像的颜色强度的空间分布信息进行统计,又可进一步分为传统的基于模型的统计方法和基于频谱分析的方法,如马尔可夫随机场模型、Fourier频谱特性等。 Statistical methods are spatial intensity distribution of the image color information statistics, statistical methods can be further divided based on the traditional model and method based on spectrum analysis, such as Markov random field model, Fourier spectrum characteristics. 结构方法首先假定纹理模式由纹理单元按照一定规则排列组成,因此纹理分析就变为确定这些单元、定量分析它们的空间排列。 Firstly, it assumes the structure of a texture pattern texture unit composed arranged according to certain rules, so texture analysis unit becomes the determination, quantitative analysis of their spatial arrangement. 本方法使用梯度图像的平均像素差平方来衡 This method uses a gradient image to the average pixel difference square value

量纹理差别。 The amount of texture differences.

[0011] 图像复杂性问题是近年来计算机视觉的研究热点问题,人眼很难从复杂场景中辨别出一个物体,但很容易从简单的场景分别出一个物体,故将缩小图像嵌入到一个相对复杂的区域是一个更好的选择。 [0011] In recent years, image complexity of the problem is the hot topic in computer vision research, human eyes is difficult to distinguish from a complex scene object, but it is easy to each object from a simple scene, so that the reduced image is embedded into a relatively complex area is a better choice.

[0012] 全局优化问题也是一个被广泛研究的问题,有许多方法被用来解决优化问题,比如模拟退火算法,蚁群算法,动态规划算法也适用于解决这一系列问题。 [0012] optimization problem is a global problem has been widely studied, there are a number of methods are used to solve optimization problems, such as simulated annealing, ant colony algorithm, dynamic programming algorithm is also applicable to solve this set of problems.

[0013] 图割算法在过去几年被广泛应用于计算机视觉领域,以解决标签问题,能量最小化问题等等,具体的应用领域包含图像分割,图像拼接,视频拼接等等,特别在在图像拼接方面,图割算法有着非常好的效果。 [0013] FIG cut algorithm is widely used in the field of computer vision over the past few years, to solve the problem of the label, like the energy minimization problem, the field of application comprises image segmentation, image mosaicing, video splicing, etc., especially in the image mosaic aspect, graph cut algorithm has a very good effect.

[0014] 由于采用本发明的技术方案,本发明以一系列图像作为输入,找到一个最佳浏览序列,形成一段不断放大的具有视觉连续性的图像浏览视频,本发明同时还具有以下有益效果: [0014] Due to the technical solution of the present invention, the present invention is a sequence of images as input, to find an optimal sequence browser, forming an enlarged section of the continuous video image browsing with visual continuity, the present invention also has the following advantages:

[0015] 1.提供了一种更为生动的图像浏览方式,使用户以一种更为生动的方式浏览图像,犹如在图像的场景中漫游。 [0015] 1. Provide a more vivid image browsing, allowing users to browse images in a more vivid way, like roaming in the scene image.

[0016] 2.对于一个旅游景点的一系列图像,本方法可生成一段较好串联该景点不同场景的视频,作为该旅游景点的宣传片。 [0016] 2. For a series of images of the tourist attractions, the process may generate a series of video period is preferably different attractions in the scene, as the trailer attractions.

[0017] 3.制作方法速度较快,不需要花费过多的人力物力去设计场景与选择图像,整个视频生成过程自动化程度高。 [0017] 3. The production method is faster, it does not need to spend too much manpower and resources to design scenarios and select the image, generate a high degree of process automation throughout the video.

附图说明 BRIEF DESCRIPTION

[0018] 图1为本发明所提供的自动生成方法的流程图。 [0018] FIG. 1 is a flowchart of an automatic generation method provided by the present invention.

[0019] 图2为本发明所提供的本地匹配步骤的子流程。 Subprocess [0019] FIG local matching step 2 of the present invention is provided.

[0020] 图3为本发明所提供的图像合成步骤的子流程图。 [0020] FIG. 3 sub-image synthesis step of the present invention provides a flow chart.

[0021] 图4为本发明所提供的自动生成方法的过程描述图,演示了将图像集作为输入, 输出最佳播放序列,播放序列中的相邻图片之间进行合成图像插值,形成输出视频的过程。 Process [0021] FIG. 4 is an automatic generation method of the present invention is provided FIG described, illustrates the input image is set as an output the best playback sequence, playback synthesized image interpolation between adjacent images in the sequence, to form an output video the process of.

[0022] 图5为本发明所提供的自动生成方法处理实例的示意简图,其在图中,第一排为按序列排列的属于一个图像播放序列中的一部分图像,第二排为生成视频中连接其中两个相邻图像的插值中间帧。 [0022] FIG. 5 automatic generation method of the present invention provides a schematic diagram of an example of a process that in the figure, the first row belonging to a part of an image sequence by the image playback sequence arranged in the second row to generate a video wherein the connection of two adjacent interpolated intermediate frame images.

具体实施方式 detailed description

[0023] 首先定义在接下来说明中要用到的符号:图像序列定义为I1, ...,In。 [0023] First define the following description to use the symbol: is defined as a sequence of images I1, ..., In. 将Ii定义为目标图像,将Ii+1的缩小版本记做pi+1。 Ii is defined as the target image, the smaller version of Ii + 1 is denoted pi + 1. Ii和pi+1的合成图像记做Si,i+1。 Ii and pi + 1 of the composite image made referred Si, i + 1. 平均像素差平方(sum squared of difference),缩写为SSD0 The average squared pixel difference (sum squared of difference), abbreviated as SSD0

[0024] 图1是本发明的基本流程图,本发明以一系列图像作为输入,以一段不断放大在图像中切换的视频作为输出。 [0024] FIG. 1 is a basic flow diagram of the present invention, the present invention is a sequence of images as input to an enlarged section of the continuous switching as an output video image.

[0025] 下面对本发明的各个流程进行详细说明: [0025] Next, each process of the present invention will be described in detail:

[0026] 1.本地匹配 [0026] 1. Local Match

[0027] 本地匹配以一系列图像作为输入(流程如图2所示),其大体流程如下:设图像中任意一幅图像为图像A,任意另一幅图像为图像B,将图像B缩小后,与图像A中的不同位置图像进行比较,计算出它们之间的图像差别,寻找所述图像A中与所述缩小的图像B之间图像差别最小的位置,将该位置作为所述缩小的图像B在图像A中的嵌入位置,该最小图像差别作为两图像最终图像差别存入图像差别表;重复上述步骤,对图像集中的图像进行两两匹配,将获得的最小图像差别存入图像差别表 [0027] In a series of local matching image as input (Scheme 2), which process substantially as follows: Let an arbitrary picture image in the image A, an image of any other image B, the image B after a reduction , with different positions in the image a, and calculates the image differences between them, to find the minimum difference between the position of the image of the reduced image of the image a and B, as the position of said reduced the embedded position in image B to image a, the difference between the minimum image difference image as a final image difference table is stored in the two images; repeating the above steps, the image of the image set pairwise matching, the minimum difference image obtained is stored in the difference image table

[0028] 图像差别包含四个方面:颜色差别,纹理差别,嵌入位置的图像复杂度,嵌入位置在图像中的位置。 [0028] The difference image contains four areas: color differences, differences in texture, image complexity is embedded position, the embedded position in the image.

[0029] HSV 空间颜色差另Ij : Δ c = α D (H) + β D (S) + γ D (V),其中α : β : y = 4:1: 1,为权重值,D(H),D(S),D(V)分别表示3个通道上的平均像素SSD [0029] HSV space color difference another Ij: Δ c = α D (H) + β D (S) + γ D (V), where α: β: y = 4: 1: 1, is the weight values, D ( H), D (S), D (V) respectively represent the average pixel SSD three channels

[0030] 纹理差别:Δ T定义为经中值滤波的两幅梯度图像的平均像素SSD [0030] The texture difference: Δ T is defined as the average of two pixel gradient by SSD median filtering of the image

[0031] 图像复杂度:使用方差V衡量目标图像区域的复杂度 [0031] The degree of image complexity: V variance measure the complexity of the target image region

[0032] 嵌入位置:对边缘的位置ρ定义惩罚值Pt(p),靠近图像中央的嵌入位置与图像边缘的嵌入位置相比较,其图像差别值小,其惩罚值Pt(P)小,对于中间的位置,该项为0。 [0032] The embedded position: position ρ defined penalty values ​​edges Pt (p), the embedded position near the center of the image is embedded position of the image edge as compared to that image difference value is small, which penalty value Pt (P) is small, for the middle position, that is zero. 也即靠近图像中央的嵌入位置给予较小的惩罚,在图像边缘的位置给予较大的惩罚。 I.e. embedded position near the center of the image given a small penalty, given a large penalty in the position of the image edge.

[0033] 比较过程按照图2所示进行,对某嵌入位置首先比较颜色差别,若颜色差别大于某一个设定阈值,则跳过该位置,去其他位置进行比较,否则继续比较纹理差别,图像复杂度,嵌入位置三个因素,直到得到该位置的综合评价。 [0033] Comparative process was conducted as shown in FIG. 2, an embedding position of the first color difference comparison, if a color difference greater than a set threshold, then skip this position, compared to other positions, or continue to compare the texture difference image complexity, the embedded position three factors until a comprehensive evaluation of the position.

[0034] 综合上述几项,将第i张图像的缩小版本(大小为r)嵌入第j张图像的消耗定义如下: [0034] The above items, the reduced version of the i-th images (size r) embedded in the j-th image sheet consumed defined as follows:

「00351 Costr(M) = mmCwjAcCip) - w2¥(p) + w3AT(i,p) Hh w4Pt(p>) LJ p €C "00351 Costr (M) = mmCwjAcCip) - w2 ¥ (p) + w3AT (i, p) Hh w4Pt (p>) LJ p € C

[0036] 其中ρ是嵌入位置,G是所有位置的集合,W1 = 0. 4,w2 = W3 = W4 = 0. 2,为权重值,其余几项为上述描述图像差别要考虑的因素,最后,考虑图像i的不同大小,将图像i贴入图像j的消耗定义如下: [0036] where ρ is the embedded position, G is the set of all positions, W1 = 0. 4, w2 = W3 = W4 = 0. 2, is the weight value, the remainder of several factors to consider the difference image described above, and finally , consider the definition of images consume different sizes i, image i paste image j as follows:

「00371 Cost(i, 0 二- min(Costr(i,j)) LJ re'R "00371 Cost (i, 0 two - min (Costr (i, j)) LJ re'R

[0038] 其中R为我们考虑的所有分辨率集合(一般选取16X 16,32X32,64X64)。 [0038] wherein R is the set we consider all resolutions (generally selected 16X 16,32X32,64X64).

[0039] 计算所有图像两两之间的Cost(i,j),最后可以得到存储所有图像之间差别的表Best Map (图像差别表)。 [0039] All calculation Cost (i, j) between every two images, the difference between the last obtained image is stored in all the tables Best Map (image difference table).

[0040] 2.全局优化 [0040] 2. Global Optimization

[0041] 全局优化以本地匹配得到的图像差别表作为输入,问题要产生一个最优播放序列,使得序列中不出现重复图像,并且相邻图像之间的图像差别之和最小,该问题通过动态规划解决,建立一个二维表格Bestkq,表格中的元素BesUeq (i,j). Ct表示长度为i的以第j个图像为结尾的序列的最小总消耗,表格同时还存储序列中前一个图像的索引,动态规划递归方程表示如下: [0041] Global optimization to match the image difference table of the local obtained as input, to produce an optimal playback problems sequence, such that no duplicate image sequence, and the difference of the image between the adjacent images and the minimum, the dynamic problem planning solutions, to create a two-dimensional table Bestkq, elements BesUeq (i, j) in the table. Ct represents the length of the j-th picture to the end of the minimum total consumption of the sequence of i, the table also stores an image in the sequence before the index, dynamic programming recursive equation is expressed as follows:

[0042] BestSeq(If)-Ct = mm (BestSeq(i - l,k).Ct + Cost0>k) + Penalty) [0042] BestSeq (If) -Ct = mm (BestSeq (i - l, k) .Ct + Cost0> k) + Penalty)

[0043] 其中Cost(j,k)是从Best Map中查得的将j贴入k的消耗,Penalty用以防止序列中重复图像的出现,若图像j已在之前的序列中出现,Penalty就是一个很大的数字,否则,Penalty 就是0。 [0043] Cost wherein (j, k) is obtained from the investigation of the Best Map paste k j is consumed, to prevent occurrence of the image Penalty repeated sequence, if j has occurred before the image sequence, is Penalty a very large number, otherwise, Penalty is 0.

[0044] 动态规划结束之后,在表格中最后一行找到总消耗最小的一个元素,通过索引信息进行回溯,可以得到一个最优序列。 After the end of [0044] dynamic programming, in the last line of the table to find the smallest element of total consumption, backtracking through the index information, you can get an optimal sequence.

[0045] 3.图像合成 [0045] The image synthesis

[0046] 在得到最优播放序列之后,对序列中前后两张图像在各个放大阶段进行合成(流程如图3所示),本发明的图像合成策略,采用图割结果进行引导,得到嵌入图像&吣和Ii 进行合成的最佳合成边界,通过该边界进行扩张,生成包含Pw中所有像素点alpha值的核K0 [0046] After the optimal sequence of playback, the image embedded in the two images before and after the sequence (Scheme 3) at various stages of amplification, the image synthesizing strategy of the present invention, using the results of FIG cutting guide, to give Qin and & Ii synthesis optimal synthesis boundaries, to expand through the boundary, generating a core comprising K0 Pw alpha value in all pixels

[0047] 图割的能量函数定义如下: [0047] FIG cut energy function defined as follows:

Figure CN101504775BD00071

[0049] 其中Edata是一阶能量,Esmooth是二阶能量。 [0049] wherein Edata is a first order energy, Esmooth second order energy. α是相对重要性,L (ρ)表示给ρ点标记的标签,1表示该点采用Pi+1的像素,O表示采用Ii的像素。 α is the relative importance, L (ρ) [rho] represents a point marking label 1 indicates that the point of using the pixel Pi + 1, O for pixel Ii employed.

[0050] 一阶能量定义如下: [0050] The first order power is defined as follows:

「00511 Ptatsd!st(p,oenter) I (ρ, 1) — center) "00511 Ptatsd st (p, oenter) I (ρ, 1) -! Center)

[0052] 其中,η是Pw的大小,设k = 0. 6,dist是计算两个像素点距离的函数,center是Pi+1的几何中心。 [0052] where, [eta] is the size of Pw, set k = 0. 6, dist two pixels is calculated from the function, center is the geometric center of Pi + 1.

[0053] 二阶能量定义如下: [0053] Second Order energy is defined as follows:

Figure CN101504775BD00072

[0055] 其中,L(a), b表示L(a)上像素b的颜色值,L(a) = O表示是目标图像,L(a) = 1 表示是缩小的图像。 [0055] wherein, L (a), b represents the color value L (a) the pixels b, L (a) = O represents a target image, L (a) = 1 is represented by the reduced image.

[0056] 使用上述定义的能量函数,采用最大流方法可以求得嵌入图像Pw和Ii进行合成的最佳合成边界,边界将图像分为A和E两个部分。 [0056] using the energy function defined above, can be obtained using the method of maximum flow embedded image Pw and optimum reaction Ii synthesis boundaries, which image is divided into two portions A and E. 通过该边界,生成包含Pw中所有像素点alpha值的核K,生成原则如下: By this boundary, it generates a core K Pw alpha value in all pixels, to generate the following principles:

[0057] •对于区域A中的像素点,给定alpha值 [0057] • For pixel area A, a given alpha value

[0058] •对于属于E中的像素点,进行4邻接扩张,给定alpha值:印(ρ) [0058] • For the pixels belonging to the E, 4 for expansion adjacent to a given alpha value: printing ([rho])

[0059] 其中,η为当前Pw大小,w为视窗大小,δ为一给定常数,st印(ρ)表示像素ρ经过扩张到达的步数。 [0059] wherein, η is the size of the current Pw, w is the window size, δ is a given constant, st printing ([rho]) [rho] represents a pixel expansion reached after several steps.

[0060] 最后得到合成图像: [0060] The finally obtained combined image:

[0061] [0061]

Figure CN101504775BD00073

[0062] 其中Ί;表示与P对应的目标图像区域。 [0062] wherein Ί; represents a target image region corresponding to the P.

[0063] 事实上有可能存在图割结果不理想的结果,若图割得到区域A小于整个区域面积的1/4,则放弃采用图割结果,转而使用高斯函数生成alpha核K : [0063] In fact there may result undesirable FIG cut results, obtained when the cut FIG region A is smaller than 1/4 of the entire area of, using the graph cut result is discarded in favor of using a Gaussian function to generate alpha nuclei K:

[0066] 其中(Xtl,y0)为P几何中心的坐标,σ x2禾口σ y2是方差,η为当前Pw大小,w为视窗大小,k用以调节核的平滑度,在此被定为4. 5。 [0066] wherein (Xtl, y0) coordinates of the geometric center P, σ x2 Wo mouth σ y2 is the variance, [eta] is the size of the current Pw, w is the window size, k to adjust the smoothness of the nucleus, where it is designated as 4.5. [0067] 4.维护帧间关联性 [0067] 4. Maintenance association inter

[0068] 在序列中,当Pw填满了视窗,合成图像由Si^1RST Si+1,i+2,在此就会产生视觉突变,为防止该突变发生,将前面的合成图像与后面的合成图像进行几个步骤的混合,加入过渡帧: [0068] In sequence, when Pw filled windows, synthesized image generated by Si ^ 1RST Si + 1, i + 2, this will produce a visual mutation, the mutation in order to prevent the front and back of the synthesized image mixing the composite image in several steps, transition frames added:

[0069] Tran = α Si, i+1+(1_ α ) Si+1, i+2 [0069] Tran = α Si, i + 1 + (1_ α) Si + 1, i + 2

[0070] 其中Tran表示过渡帧,α的初始值设定为0. 9,每一部放大将其值减小0. 025,直至其值减小至0。 [0070] wherein Tran represents a transition frame, the initial value of α is set to 0.9, which is an enlarged portion of each value is reduced 0.025, until it is reduced to the value 0.

[0071] 5.生成视频 [0071] The generated video

[0072] 经过上述几个过程,在每个放大步骤生成视频的一帧,所有这些帧组成了输出的放大视频。 [0072] After the above-described several processes, each generating a video in an amplification step, all of the video frames of the amplified output. 图5展示了其中部分连接相邻图像的帧。 Figure 5 shows the frame in which the connecting adjacent portions of the image.

Claims (10)

  1. 1. 一种基于图像集的漫游视频自动生成方法,其特征在于该方法包括以下步骤:(1)本地匹配:设图像集中任意一幅图像为图像A,任意另一幅图像为图像B,将图像B 缩小后,与图像A中的不同位置图像进行比较,计算出它们之间的图像差别,寻找所述图像A中与缩小的图像B之间图像差别最小的位置,将该位置作为所述缩小的图像B在图像A中的嵌入位置,最小图像差别作为两图像最终图像差别存入图像差别表;重复上述步骤,对图像集中的图像进行两两匹配,将获得的最小图像差别存入图像差别表;(2)全局优化:根据本地匹配得到的图像差别表,找出图像差别总和最小的图像排列序列,作为播放序列,其中,嵌入的图像排在被嵌入的图像的后面;(3)图像合成:按照得到的播放序列,按前后次序,设序列中原始图像为i、i+1、 i+2……i+(nl)、i+n,将i+1缩小并 An automatic roaming video image is generated based on the set, characterized in that the method comprises the steps of: (1) Local Match: set any image set as an image of image A, the image of any other image B, and after the reduced image B, with different positions in the image a, and calculates the image differences between them, to find the minimum difference between the position of the image and the reduced image of image a and B, as the position of the B embedding a reduced image position in the image a, the final image as the minimum image difference image into two difference image difference table; repeating the above steps, the image of the image set pairwise matching, the minimum difference image stored in the image obtained table difference; (2) global optimization: the image obtained by matching different local table, find the smallest difference in the sum of the image sequence image arrangement, as the playback sequence, wherein the embedded image is embedded at the back of the image; (3) image synthesis: according to the obtained playback sequence, according to the order of the front and rear, arranged in the sequence of the original image i, i + 1, i + 2 ...... i + (nl), i + n, i + 1 will be reduced and 照本地匹配中获得的嵌入位置嵌入到i中,放大i和i+Ι的合成图像,直至i+Ι充满视窗;重复上述过程,进行i+Ι与i+2直至i+(nl)与i+n的处理;每个放大阶段的合成图像都作为前后原始图像之间的插值图像,所有图像输出视频的帧;最后,将所有帧连接成视频输出。 Embedding position according to the local matching obtained is embedded into i, amplifying i and the synthesized image i + iota until i + iota filled windows; repeat the process, for i + iota and i + 2 until i + (NL) and i + n processing; synthetic image are each amplification stage as the interpolation image between before and after the original image, all the images of the output video frame; Finally, all frames into a video output connector.
  2. 2.根据权利要求1所述的一种基于图像集的漫游视频自动生成方法,其特征在于所述图像差别包含2个方面:颜色差别,纹理差别;其中,颜色差别使用HSV颜色空间上的平均像素差平方来衡量;纹理差别使用梯度图像的平均像素差平方来衡量。 According to one of the claims 1 to set the roaming video image automatic generation method based on a difference image wherein said two aspects comprises: color differences, texture difference; wherein an average color differences on the HSV color space squared pixel difference measure; texture difference average squared gradient of pixel difference image to measure.
  3. 3.根据权利要求2所述的一种基于图像集的漫游视频自动生成方法,其特征在于所述图像差别包含2个方面:嵌入位置的图像复杂度,嵌入位置在图像中的位置;图像复杂度使用方差来衡量,嵌入位置的图像复杂度越大,图像差别值越小;靠近图像中央的嵌入位置与图像边缘的嵌入位置相比较,其图像差别值小。 The automatic generation of one of said two video image based on roaming set, wherein the image comprises two different aspects claim: embedded image complexity of the position, the position of the embedded position in the image; complex image the greater the degree of variance measure, the complexity of the embedding position of the image, the smaller the value of the difference image; fitted position near the center of the image compared with the position of the embedded edge of the image, a small image which is the difference value.
  4. 4.根据权利要求2或3所述的一种基于图像集的漫游视频自动生成方法,其特征在于在计算所述图像差别时,首先计算颜色差别,若颜色差别大于给定阈值则放弃该图像差别比较,否则,继续计算剩下几个因素进行完全比较。 4. In accordance with one 2 or 3 is automatically generated based on roaming video image set, wherein in the calculation of the difference image, the color difference is first calculated, if the color difference is greater than a given threshold to discard the image as claimed in claim Compare the difference, otherwise, the remaining number of factors continue to calculate full comparison.
  5. 5.根据权利要求1所述的一种基于图像集的漫游视频自动生成方法,其特征在于在步骤(1)中,将所述图像A划分为与所述缩小的图像B大小相同的不相交子区域,对所述缩小的图像B与所述子区域进行图像差别比较。 1, according to one of the automatically generated video image set based on the roaming of the preceding claims, characterized in that in step (1), the image A is divided into the same size of the reduced image B do not intersect sub-region, the reduced image of image B compared with the difference between the sub-regions.
  6. 6.根据权利要求1所述的一种基于图像集的漫游视频自动生成方法,其特征在于在步骤O)中,使用动态规划方法得到所述播放序列;设播放序列的相邻前后的两个图像中,前一个图像为被嵌入图像,后一个图像的缩小图像为嵌入图像,在输出信息中,还包括播放序列中的嵌入图像的最佳初始缩小尺寸信息以及被嵌入图像的被嵌入位置信息。 1, according to one of the automatically generated video image set based on the roaming of the preceding claims, characterized in that in step O), the method using dynamic programming to give the play sequence; provided two adjacent sequences before and after the playback image, the first image as being embedded image, a reduced image after the image is embedded into an image in the output information, further comprising optimum initial embedded image playback sequence information is reduced in size embedded image position information is embedded .
  7. 7.根据权利要求1所述的一种基于图像集的漫游视频自动生成方法,其特征在于设播放序列的相邻前后的两个图像中,前一个图像为被嵌入图像,后一个图像的缩小图像为嵌入图像,在步骤(3)中,对每个放大阶段的合成图像中的嵌入图像和被嵌入图像进行图像融合处理。 1, according to one of the automatically generated based on roaming video image set, wherein two adjacent images before and after the set playback sequences, as the previous image is an embedded image, a reduced image of the claims the image is embedded into an image in step (3), the composite image embedded in each amplifying stage are embedded image and image fusion process.
  8. 8.根据权利要求6所述的一种基于图像集的漫游视频自动生成方法,其特征在于所述图像融合处理采用以下处理方法:它采用图割技术,获得嵌入图像和被嵌入图像的嵌入目标区域融合的能量最小边界;在得到能量最小边界的条件下,对边界内嵌入图像的像素点给予初始alpha值,在边界上做4邻接扩张,每步扩张得到的像素点alpha值线性递减;最终得到嵌入图像的alpha值的核K ;使用K,将嵌入图像与被嵌入图像的嵌入目标区域进行alpha颜色混合,得到合成图像。 According to claim 6, wherein one of the automatically generated video image set based on the roaming process, wherein said image processing method fusion following: It uses graph cut technique, and the obtained embedded image the embedding object embedded image energy minimum boundary area integration; minimal condition boundary resulting energy given to the pixels of the embedded image within the boundaries of the initial alpha values ​​do 4 adjacent to the expansion at the boundary, pixel alpha value linearly each step expansion obtained decremented; final embedded image obtained kernel K alpha value; using K, embedded with the embedding target area image embedded image are alpha color mixing, to obtain a composite image.
  9. 9.根据权利要求1所述的一种基于图像集的漫游视频自动生成方法,其特征在于合成图像作为视频帧输出,所述方法在合成视频时维护其帧间连续性。 According to claim 1, wherein one of the automatically generated based on roaming video image set, wherein the composite image is outputted as a video frame, the method of maintaining the continuity in the synthesis of the video frames.
  10. 10.根据权利要求9所述的一种基于图像集的漫游视频自动生成方法,其特征在于所述维护帧间连续性步骤采用保留之前合成图像与当前合成图像进行alpha混合的方式。 10. A method according to claim 9, said automatically generated based on roaming video image set, wherein the step of maintaining the continuity of the inter mode using alpha blending the composite image combined image before the current reservation.
CN 200910096855 2009-03-19 2009-03-19 Roaming video automatic generation method based on image set CN101504775B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN 200910096855 CN101504775B (en) 2009-03-19 2009-03-19 Roaming video automatic generation method based on image set

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN 200910096855 CN101504775B (en) 2009-03-19 2009-03-19 Roaming video automatic generation method based on image set

Publications (2)

Publication Number Publication Date
CN101504775A true CN101504775A (en) 2009-08-12
CN101504775B true CN101504775B (en) 2011-08-31

Family

ID=40977010

Family Applications (1)

Application Number Title Priority Date Filing Date
CN 200910096855 CN101504775B (en) 2009-03-19 2009-03-19 Roaming video automatic generation method based on image set

Country Status (1)

Country Link
CN (1) CN101504775B (en)

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20040179262A1 (en) 2002-11-25 2004-09-16 Dynamic Digital Depth Research Pty Ltd Open GL
CN1662933A (en) 2002-05-24 2005-08-31 德耐皮克斯情报图象公司 Method and apparatus for comprehensive and multi-scale 3D image documentation and navigation
WO2007016596A2 (en) 2005-07-29 2007-02-08 Pamela Barber Digital imaging method and apparatus
US20070150188A1 (en) 2005-05-27 2007-06-28 Outland Research, Llc First-person video-based travel planning system

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN1662933A (en) 2002-05-24 2005-08-31 德耐皮克斯情报图象公司 Method and apparatus for comprehensive and multi-scale 3D image documentation and navigation
US20040179262A1 (en) 2002-11-25 2004-09-16 Dynamic Digital Depth Research Pty Ltd Open GL
US20070150188A1 (en) 2005-05-27 2007-06-28 Outland Research, Llc First-person video-based travel planning system
WO2007016596A2 (en) 2005-07-29 2007-02-08 Pamela Barber Digital imaging method and apparatus

Also Published As

Publication number Publication date Type
CN101504775A (en) 2009-08-12 application

Similar Documents

Publication Publication Date Title
Park et al. High quality depth map upsampling for 3d-tof cameras
Zheng et al. Learning based digital matting
US20030223622A1 (en) Method and system for enhancing portrait images
Zhang et al. Synthesis of progressively-variant textures on arbitrary surfaces
US7636485B2 (en) Method and system for enhancing portrait images that are processed in a batch mode
US20020118875A1 (en) Image extraction from complex scenes in digital video
US20130094780A1 (en) Replacement of a Person or Object in an Image
Unger et al. TVSeg-Interactive Total Variation Based Image Segmentation.
US20070146360A1 (en) System And Method For Generating 3D Scenes
Wang et al. Video tooning
Li et al. Video object cut and paste
Hillman et al. Alpha channel estimation in high resolution images and image sequences
Liao et al. Automatic caricature generation by analyzing facial features
Brooks et al. Self-similarity based texture editing
Imelińska et al. Semi-automated color segmentation of anatomical tissue
US20130229581A1 (en) Juxtaposing still and dynamic imagery for cliplet creation
CN101826217A (en) Rapid generation method for facial animation
CN101661613A (en) Image restoration method based on image segmentation, and system therefor
Reinhard et al. Calibrated image appearance reproduction
Bleyer et al. A stereo approach that handles the matting problem via image warping
Wang et al. Simultaneous matting and compositing
CN101588459A (en) Video keying processing method
Inamoto et al. Intermediate view generation of soccer scene from multiple videos
CN101777180A (en) Complex background real-time alternating method based on background modeling and energy minimization
Zhang et al. Exemplar-based image inpainting using color distribution analysis

Legal Events

Date Code Title Description
C06 Publication
C10 Request of examination as to substance
C14 Granted