CN101477695A - Method for fast painting image and video - Google Patents

Abstract

A method for rapid coloring images and videos relates to a method for coloring a gray image and a gray video into a colored image and a colored video, as well as re-coloring a colored image and a colored video. The method comprises the following steps: a small amount of color information is sketched by a user on the gray image; the local connection energy of a pixel is defined on the basis of the edge analysis and the analysis of texture pattern; the two-dimensional dynamic programming algorithm is adopted for calculating the minimum distance from the pixel to all known color points; and the distance is taken as a weight by which colors are mixed. The coloring method provided by the invention is rapid, and can enable the user to obtain a coloring effect picture in a real-time manner, thereby realizing interactive editing of the coloring effect picture. Due to a changing gradient of a color on a time axis, coloring is extended to the video. The re-coloring of the colored image can be realized by comprehensive constraint of color and brightness.

Description

The method of a kind of image and video fast colorizing

Technical field

The present invention relates to image and video fast colorizing method, what relate in particular to gray level image and video is colored as coloured image and screen and the method for restaining of coloured image and video.Belong to the Image Information Processing field.

Background technology

Can improve visual attraction greatly for interpolation colouring informations such as black-and-white photograph, classic black film, science data figure, also be video display worker's active demand for existing coloured image and restaining of video in addition.Painted (Colorization) is that Wilson Markle proposed in 1970, be used at first film industry, now making a general reference colouring information is added on still image and the dynamic video, is the process that a colouring information with three-dimensional adds the image or the video (being continuous a series of images) of the monochrome information of having only one dimension to.Therefore because different colors may show same brightness, the process of extrapolating value of color by gray level image is an ill-conditioning problem (ill-posed problem), the correct scheme that neither one is absolute.

Colorize method mainly is divided into two big classes: semi-automatic (Semi-automatic) is painted and user's guiding (User-guided) is painted.Relation between semi-automatic painted colouring information by the existing image library of analytic learning and brightness, the texture, with this relationship map to new gray level image, realization gray-scale map painted.The benefit of this class colorize method is exactly the participation that does not need the people, can carry out painted to image, video in batches, but semi-automatic colorize method is quite high for the requirement of image library, and need carry out analytic learning to various situations, and often the result of Shi Xianing can not be satisfactory.The user guides the artificial interpolation part colouring information of painted needs, by computing machine other parts is carried out color again and fills.Because visual effect itself just has subjectivity, because people's participation strengthens painted constraint, can be more near expected effect.The painted main flow research direction that become of user's guiding.

Painted the biggest problem is exactly the big problem of calculated amount at present.Usually for computing time of the still image of a secondary 320*240 all more than 10 seconds, the user is difficult to the mutual painted result to image and makes amendment.Painted speed for video is slow especially.And the border color of existing colouring algorithm to obscure (color confusion) phenomenon very serious.

Summary of the invention

Technical matters: the objective of the invention is to the deficiency at existing dye technology: problems such as speed is slow, border color is obscured, interactivity difference provide a kind of real-time, interactive, brightness and texture pattern analysis-by-synthesis, add border and a kind of image of gradient constraint and the method for video fast colorizing.

Technical scheme: the present invention at first is mapped to three-dimensional color space with gray level image, and coloring problem is extrapolated unknown colouring information just by known brightness value in the image,

Following three hypothesis are at first proposed:

◆ the color of pixel correlativity of same area is big, and the color correlation of zones of different is little,

◆ the color of pixel of the similar brightness of the same area is also similar,

◆ change color is level and smooth;

As follows for the painted step of still image:

Steps A is added a little colouring information alternately with paintbrush tool on gray level image;

The Luminance Distribution of step B analysis-by-synthesis gray level image, texture distribute and edge of image distributes, and calculate the influence degree that is subjected to known color of the gray-scale pixels in the image;

Step C generates colour picture by color is carried out hybrid rending;

Step D user is as dissatisfied to partial result, can add again or revises colouring information by paintbrush tool, gets back to steps A, until obtaining satisfactory result.

Carry out painted for video (being continuous a series of images), continuous fragment at Same Scene, consider that brightness is similar in the variation of time shaft with color component, by a frame is wherein carried out mutual painted after, according to the variation of color at time shaft, with colouring information to adding other successive frames to.

Coloured image is carried out painted, only use and the Luminance Distribution among the step B is replaced known color distribution get final product.

Carrying out painted, traditional method for video (being continuous a series of images) is to utilize light stream to calculate, and inefficiency can't reach real-time.The present invention is directed to the continuous fragment of Same Scene, consider brightness and color component variation similarity at time shaft, by a frame is wherein carried out mutual painted after, according to the brightness intensity of variation colouring information is added to other successive frames, improved the painted efficient of video greatly.

Dissatisfied as the user to the color in coloured image or the video, its color is carried out part or whole process of replacing, be called restaining of coloured image.Only use the brightness analysis among the above-mentioned steps B is replaced with known color analysis, just painted calculation method can be applied on the restaining of coloured image.

Beneficial effect: compared with prior art, it is painted and restain method to the present invention proposes a kind of image of real-time, interactive and video.Solved that the classic method color is obscured, coloring process complexity, problem that calculated amount is big.For film industry, particularly the editor of cartoon is painted has very important significance.

Description of drawings

Fig. 1 is the entire block diagram of the colorize method of image of the present invention and video;

Embodiment

As shown in Figure 1, the input of colouring algorithm of the present invention can be a gray level image, coloured image or video data, need carry out scene earlier for video data cuts apart, and then the painted processing that the first two field picture of the continuous fragment of Same Scene is carried out, the user can interactivity the interpolation color pencil be drawn on the image, painted module jointing edge constraint and texture pattern analysis, there emerged a the shortest path of pixel by energy definition calculating to color dot, carry out the color value that blend of colors calculates each pixel then, and show in real time, as dissatisfied, can delineate or revise color again, till drawing result to local result.Also to utilize brightness to extrapolate color gradient for video is painted, carry out painted residue frame in the gradient of time shaft.

For ease of profound understanding technology contents of the present invention, the present invention is described in detail below in conjunction with accompanying drawing.

1. to the video data of input, carry out the successive frame contrast, think that when difference value surpasses threshold value scene changes.Continuous fragment for Same Scene is carried out painted respectively.At first extracting first two field picture carries out painted.

2. the user delineates out a little colored lines (seeing the intermediate data figure of Fig. 2) by paintbrush on gray-scale map.Painted target is exactly Luminance Distribution and the texture pattern by the analysis gray-scale map, and a little colouring information that utilizes the user to provide is extrapolated the color of entire image.Just find out one or several colour element that influences the gray-scale pixels maximum, carry out blend of colors again.We with its abstract be a shortest path minimum energy problem.The key problem of shortest path is a local link energy problem definition.We define: and J (p, q)=w _GJ _G(q)+w _TJ _T(p, q),

Wherein J is neighbor pixel p, the energy of q, J _TBe texture pattern component, w _EBe edge component weight, w _GBe gradient component weight, w _TBe texture pattern component weight.

3. edge extracting and analysis.In order to solve the problem that traditional stains method color is obscured, particularly to cartoon image, in local energy definition, we have added edge component, and adjust by weighted value the proportion of edge component: J ' (p, q)=w _EJ _E(p, q)+J ' (p, q)=w _EJ _E(p, q)+w _GJ _G(q)+w _TJ _T(p, q), J wherein _EBe edge component, J _GBe the gradient component.By edge extracting operator extraction edge pixel, and edge strength analyzed, passing through of color played attenuation.Can effectively prevent the generation of the phenomenon that color is obscured.

4. texture pattern analysis.Often exist a large amount of repetition textures in the image, extract, color is filled fracture and inaccurate occurs if be used as the edge merely.By the texture pattern analysis, find out the texture continuum, reduce this regional texture pattern component J _T,, strengthen texture pattern component J in the part of texture variations _TSimultaneously can strengthen texture pattern component weight w for the more image of repetition textures such as cartoon image _T

5. according to the definition of above energy, the pixel of the connection energy that calculates constitutes a figure as node, obtains the bee-line of the pixel that each pixel delineates to the user.

6. blend of colors, the color that the user delineates is a kind of incessantly, and so every kind of color all can exert an influence to other pixels, and we mix then according to the bee-line to pixel of the every kind of color value that calculates:

$C=\frac{\mathrm{\Σw}\left(d_c\right)c}{\mathrm{\Σw}\left(d_c\right)},$

Wherein C represents color of pixel value to be asked, d _cFor the user of color c delineates pixel to the bee-line of waiting to ask pixel.W (.) hybrid weight, we define $w\left(r\right)=e^{{-\mathrm{\τ}}^2/2}.$ In order to accelerate blend of colors speed, we only mix with the shortest front three of computed range.

7. the painted time shaft analysis of screen.When scene identical, and motion consecutive hours, our hypothesis color on time shaft also has continuity, adds the colouring information of painted first two field picture to other frame:

$\frac{\∂Y}{\∂t}=\frac{\∂U}{\∂t}=\frac{\∂V}{\∂t},$

Wherein t is the time, and Y, U, V are three components of YUV color space, and Y represents brightness, and U represents the first dimension colourity, and V represents the second dimension colourity, and the difference between Y and the V becomes aberration.

8. coloured image is restained.The user delineates a little colored lines in all the other of hope change color.Coloured image not only provides monochrome information, and colouring information also is provided.Colouring algorithm is added the color constraint, make and restain algorithm more fast with effective.Energy definition, shortest path and blend of colors are constant.

Claims (2)

1. the method for image and video fast colorizing is characterized in that at first gray level image is mapped to three-dimensional color space, coloring problem just becomes by known brightness value, extrapolates color component,

Following three hypothesis are at first proposed:

◆ the color of pixel correlativity of same area is big, and the color correlation of zones of different is little,

◆ the color of pixel of the similar brightness of the same area is also similar,

◆ change color is level and smooth;

As follows for the painted step of still image:

Steps A is added a little colouring information alternately with paintbrush tool on gray level image;

The Luminance Distribution of step B analysis-by-synthesis gray level image, texture distribute and edge of image distributes, and calculate the influence degree that is subjected to known color of the gray-scale pixels in the image;

Step C generates colour picture by color is carried out hybrid rending;

Step D user is as dissatisfied to partial result, can add again or revises colouring information by paintbrush tool, gets back to steps A, until obtaining satisfactory result.

Carry out painted for video, at the continuous fragment of Same Scene, consider that brightness is similar in the variation of time shaft with color component, by a frame is wherein carried out mutual painted after, according to the variation of color at time shaft, with colouring information to adding other successive frames to.

Dissatisfied as the user to the color in coloured image or the video, its color is carried out part or whole process of replacing, be called restaining of coloured image.Only use the brightness analysis among the above-mentioned steps B is replaced with known color analysis, just painted calculation method can be applied on the restaining of coloured image.

2. the method for image according to claim 1 and video fast colorizing is characterized in that, coloured image is carried out painted, only uses and the Luminance Distribution among the step B is replaced known color distribution gets final product.

Images