CN101865859B - Detection method and device for image scratch - Google Patents

Abstract

The embodiment of the invention discloses a detection method and a detection device for an image scratch. The method comprises the following steps of: projecting a single-color component in a two-dimensional image along a scratch direction to obtain the sum of single-dimensional scratch projection; calculating a filter function of pixel points according to a complex wavelet function; performing convolution on the sum of the filter function and the single-dimensional scratch projection to obtain a single-color multi-ridgelet coefficient, and calculating a fused multi-ridgelet coefficient of the image according to the single-color multi-ridgelet coefficient; and determining the edge of the scratch according to an imaginary part of the fused multi-ridgelet coefficient. The image is subjected to multi-ridgelet conversion so as to obtain the multi-ridgelet coefficient, and the precision of positioning the scratch can be improved by adopting the multi-ridgelet coefficient.

Description

A kind of detection method of image scratch and device

Technical field

The present invention relates to image processing field, relate in particular to a kind of detection method and device of image scratch.

Background technology

Along with the rise of digital television techniques, after being digitized, a large amount of film movies passes through media play.Since cinefilm through long-term storage with repeatedly play, exist the zone that degrades of types such as dust, mildew, cut, it is increasingly important therefore to detect and repair the film regional digital technology that degrades automatically.Wherein, The film cut is meant the silver granuel peeling phenomenon that wire distributes in the cinefilm; Normally by hard particle in that to be parallel on the direction of motion of cinefilm scraping caused, very little although the width of cut (about 1～5 pixel) is compared whole image, run through whole image often; Cause the discontinuous of video, influence the play quality of video image.Because during the film cut and occur in the adjacent locations of several two field pictures that link to each other, the time and completely random, so the detection of film cut is challenging.

Detect to the image scratch of film, existing Ridgelet (ridge ripple) conversion is applicable to the detection of straight line, therefore can be used for the cut of detected image, and present scratch detection all is to adopt real number Ridgelet conversion.

In realizing process of the present invention, the inventor finds:

The scheme of prior art can't accurately be oriented the scope of scored area, and promptly the degree of accuracy of the scope of the location scored area of prior art is very low.

Summary of the invention

The embodiment of the invention provides a kind of detection method and device of image scratch, to improve the degree of accuracy of location cut.

According to the one side of the embodiment of the invention, a kind of detection method of image scratch is provided, said method comprises:

Single color component in the two dimensional image is done projection along the cut direction, obtain the projection of one dimension cut with;

Filter function according to multiple wavelet function calculating pixel point; The real part of said multiple wavelet function adopts the second derivative form of Gaussian function; Be used for the matching detection cut, the imaginary part of said multiple wavelet function is the first order derivative of Gaussian function, is used to locate the edge of cut;

With said filter function and the projection of said one dimension cut with make convolution, obtain the multiple ridge wave system number of single color, count the multiple ridge wave system number of fusion of computed image according to the multiple ridge wave system of said single color;

Confirm the edge of cut according to the imaginary part of the multiple ridge wave system number of said fusion.

According to the embodiment of the invention on the other hand, a kind of pick-up unit of image scratch is provided, said device comprises:

Projection module is done projection with the single color component in the two dimensional image along the cut direction, obtain the projection of one dimension cut with;

Computing module; According to the filter function of multiple wavelet function calculating pixel point, the real part of said multiple wavelet function adopts the second derivative form of Gaussian function, is used for the matching detection cut; The imaginary part of said multiple wavelet function is the first order derivative of Gaussian function, is used to locate the edge of cut;

Processing module, with said filter function and the projection of said one dimension cut with make convolution, obtain the multiple ridge wave system number of single color, count the multiple ridge wave system number of fusion of computed image according to the multiple ridge wave system of said single color;

Operational module is confirmed the edge of cut according to the imaginary part of the multiple ridge wave system number of said fusion.

The embodiment of the invention obtains multiple Ridgelet coefficient through image being done plural Ridgelet conversion, confirms the edge of cut according to the imaginary part of the multiple ridge wave system number of said fusion, thereby can improve the degree of accuracy of location cut.

Description of drawings

In order to be illustrated more clearly in the embodiment of the invention or technical scheme of the prior art; The accompanying drawing of required use is done to introduce simply in will describing embodiment below; Obviously, the accompanying drawing in describing below only is some embodiments of the present invention, for those of ordinary skills; Under the prerequisite of not paying creative work property, can also obtain other accompanying drawing according to these accompanying drawings.

Fig. 1 is the detection method process flow diagram of the image scratch that provides of the embodiment of the invention one;

Fig. 2 is the synoptic diagram of the image scratch that provides of the embodiment of the invention two;

Fig. 3 is the detection method process flow diagram of the image scratch that provides of the embodiment of the invention two;

Fig. 4 is the detection method process flow diagram of the image scratch that provides of the embodiment of the invention three;

Fig. 5 is the method flow diagram of the false cut of the eliminating that provides of the embodiment of the invention four;

Fig. 6 is the detection method process flow diagram of the image scratch that provides of the embodiment of the invention five;

Fig. 7 be the embodiment of the invention five provide cut apart after overlapping subgraph piecemeal synoptic diagram;

Fig. 8 is the pick-up unit structural drawing of the image scratch that provides of the embodiment of the invention six;

Fig. 9 is the pick-up unit structural drawing of the another image scratch that provides of the embodiment of the invention six;

Figure 10 is the pick-up unit structural drawing of the image scratch again that provides of the embodiment of the invention six.

Embodiment

To combine the accompanying drawing in the embodiment of the invention below, the technical scheme in the embodiment of the invention is carried out clear, intactly description, obviously, described embodiment only is the present invention's part embodiment, rather than whole embodiment.Based on the embodiment among the present invention, those of ordinary skills are not making the every other embodiment that is obtained under the creative work prerequisite, all belong to the scope of the present invention's protection.

For making technical scheme of the present invention and advantage clearer, will combine accompanying drawing that embodiment of the present invention is done to describe in detail further below.

Embodiment one

As shown in Figure 1, the detection method process flow diagram of the image scratch that provides for the embodiment of the invention, this method comprises:

101: the single color component in the two dimensional image is done projection along the cut direction, obtain the projection of one dimension cut with;

102: according to the filter function of multiple wavelet function calculating pixel point;

103: with filter function and the projection of one dimension cut with make convolution, obtain the multiple ridge wave system number of single color, count the multiple ridge wave system number of fusion of computed image according to the multiple ridge wave system of single color;

104: the edge of confirming cut according to the imaginary part that merges multiple ridge wave system number.

The method that provides through present embodiment; Image is done plural Ridgelet conversion obtain multiple Ridgelet coefficient; Wherein, the position and the intensity of the Gray Level Jump of the imaginary part extreme point correspondence image of multiple Ridgelet coefficient can be used for the cut edge and the vertical location of image border; And the phase information of the multiple Ridgelet coefficient that combination imaginary part coefficient and real part coefficient obtain then can be used to distinguish the edge of vertical cut and vertical step, thereby finally can improve the degree of accuracy of location cut.

Embodiment two

In practical application because image scratch shows as near normal (angle of inclination＜5 °), so present embodiment with image scratch be vertical be example explanation, for cut, handle through the method in the subsequent implementation example with certain inclination angle.

In most cases, image scratch all fully can be approximate with vertical line.As shown in Figure 2, be the synoptic diagram of image scratch.Among Fig. 2, the discrete levels coordinate of n presentation video, the discrete vertical coordinate of m presentation video, image scratch is vertical, shows as the vertical line perpendicular to the n axle.Wherein, the value of n and m is integer or natural number.

Because coloured image can be handled through dividing three kinds of basic colors; Therefore the embodiment of the invention is divided into the example explanation with what image is carried out R (red), G (green), B (indigo plant) three primary colours; It is understandable that, can be other three kinds of basic colors with image division also.

Present embodiment to R, G, the B color component of image do respectively 0 the degree direction ridge small echo (Ridgelet) conversion, obtain corresponding Ridgelet conversion coefficient

S=s wherein ₁..., s _JBe the parameter in the conversion process, represent a series of decomposition scale; X is an independent variable, does not have the physics implication here, in ridge wavelet transformation process, is appreciated that to be horizontal coordinate; Detect the edge of cut, differentiation cut and cut through the Ridgelet conversion coefficient.

As shown in Figure 3, the method for the detected image cut that provides for present embodiment, this method comprises:

301: with the single color component f in the Two-dimensional Color Image (m n) vertically does projection, obtain the one dimension vertical projection with $\mathrm{Pf}\left(n\right)=\underset{m\∈N}{\mathrm{\Σ}}f(m,n).$

Wherein, be reference with coordinate shown in Figure 2, n is the horizontal component coordinate, m is the vertical component coordinate, because image scratch is perpendicular to the n axle, therefore image is vertically done projection promptly does projection along the cut direction; (m n) refers to the image that certain color component of coloured image constitutes to single color component f, and here, (m n) is specially the image value of single color component to f, is scalar.

With image f (m; N) vertically the direction of n axle (promptly perpendicular to) done projection; Obtain

and represent image f (m; N) pixel value of each row is added up, obtain an one dimension vertical projection with; Wherein, m, n are discrete independent variable, and value is integer or natural number; P representes the variable of projection vertically; N representes natural number.

302: obtain the multiple wavelet function that is used to detect cut according to pre-set criteria.

Wherein, the embodiment of the invention adopts the Ridgelet transform method detected image cut of plural number, need obtain the multiple wavelet function of s decomposition scale that is used to detect cut, s=s ₁..., s _JBe a series of decomposition scale parameters, its experience width range (it is generally acknowledged 1～5 pixel) decision by cut.Detailed process is following:

Plural number Ridgelet function shape such as ψ ^(r)((xcos θ+ysin θ)/s)+j ψ ⁽ⁱ⁾((xcos θ+ysin θ)/s), wherein ψ ^(r)(t), ψ ⁽ⁱ⁾(t) correspond to 1 of reality, imaginary part respectively and tie up wavelet function, θ and s are respectively direction and decomposition scale parameter, and x, y and t are continuous independent variable, and value is a real number, does not have special implication.In the present embodiment, because image scratch shows as near normal, so the θ value is 0 degree.Adopt the Ridgelet conversion of plural number, key is the selection to multiple wavelet function.The embodiment of the invention is obtained multiple wavelet function according to following pre-set criteria:

1) real part ψ ^(r)(t) the second derivative form of employing Gaussian function;

Because Gaussian function (meets Cexp ((t/s) ²) gang's function, C wherein, s is a constant, t is an independent variable) gray scale (image is made up of a series of gray-scale values) saltus step that causes of second derivative form and cut itself approaching, so the second derivative of Gaussian function can be used for scratch detection; So when confirming again wavelet function, real part ψ ^(r)(t) the second derivative form of employing Gaussian function.

2) fixing real part ψ ^(r)(t) after, select imaginary part ψ according to following pre-set criteria ⁽ⁱ⁾(t):

A) ψ ⁽ⁱ⁾(t) and ψ ^(r)(t) should be similar to formation Hilbert (Hilbert) transfer pair;

B) ψ ⁽ⁱ⁾(t) has only a zero crossing;

C) ψ ^(r)(t)+j ψ ⁽ⁱ⁾(t) envelope should have slickness and not have concussion property;

D) energy of envelope should have centrality.

Wherein, at first select imaginary part ψ ⁽ⁱ⁾(t) for having free parameter (α, the ψ of family of functions β) ^(j)(t)=sign (t) C _i(β) | t/ β | ^αExp ((t/ β) ²/ 2); Wherein, sign (t) is a sign function, C _i(β) be normaliztion constant, to guarantee || ψ ^(r)(t) || ₂With || ψ ⁽ⁱ⁾(t) || ₂Equate, || ψ ^(r)(t) || ₂Expression ψ ^(r)(t) second order norm, || ψ ⁽ⁱ⁾(t) || ₂Expression ψ ⁽ⁱ⁾(t) second order norm; Because this family's function and ψ ^(r)(t) approximate formation Hilbert transfer pair, thus satisfy above-mentioned optimization criterion a); And this family's function has only a zero crossing, therefore satisfies above-mentioned optimization criterion b); Secondly, make envelope A (t)=(ψ ^(r)(t)) ²+ (ψ ⁽ⁱ⁾(t)) ², wherein, the independent variable of t representative function does not have special implication; According to above-mentioned optimization criterion c), d), set up and the optimization aim function $\underset{\mathrm{\&alpha;},\mathrm{\&beta;}}{\mathrm{Arg}\mathrm{Min}}\mathrm{\&lambda;}\&CenterDot;{\&Integral;}_0^{+\&infin;}{\left(A^{\&prime;\&prime;}\left(t\right)\right)}^2\mathrm{Dt}+(1-\mathrm{\&lambda;})\&CenterDot;{\&Integral;}_0^{+\&infin;}{\mathrm{<figure-callout\; id="2"\; label="t"\; filenames="H200910133800XE00042.png"\; state="\{\{state\}\}">t</figure-callout>}}^2\&CenterDot;A\left(t\right)\mathrm{Dt},$ Wherein, λ is the weighting parameters of objective function, and span is 0～1, and (α β), thereby can confirm ψ to the parameter that gains freedom ⁽ⁱ⁾(t), get α=1 in the reality, β=0.83.

The embodiment of the invention is according to above-mentioned 1), 2) content, can confirm that the multiple wavelet function of s decomposition scale is:

$\left\{\begin{array}{c}{\mathrm{\&psi;}}_s\left(t\right)={\mathrm{\&psi;}}_s^{\left(r\right)}\left(t\right)+j\&CenterDot;{\mathrm{\&psi;}}_s^{\left(i\right)}\left(t\right)\\ {\mathrm{\&psi;}}_s^{\left(r\right)}\left(t\right)=s^{-1}{\mathrm{\&psi;}}^{\left(r\right)}(t/s)=s^{-1}\sqrt{\frac{2}{3\mathrm{\&beta;}}}({\left(\frac{t}{s}\right)}^2-1)\exp (-\frac{1}{2}{\left(\frac{t}{s}\right)}^2),t\&Element;R,s\&Element;R^{+}\\ {\mathrm{\&psi;}}_s^{\left(i\right)}\left(t\right)=s^{-1}{\mathrm{\&psi;}}^{\left(i\right)}(t/s)={\left(\mathrm{s\&beta;}\right)}^{-1}\&CenterDot;\left(\frac{t}{\mathrm{s\&beta;}}\right)\&CenterDot;\exp (-\frac{1}{2}{\left(\frac{t}{\mathrm{s\&beta;}}\right)}^2)\end{array}\right.---\left(1\right)$

Wherein, ψ _s(t) the multiple wavelet function of s decomposition scale that adopts for the embodiment of the invention,

With

Be respectively ψ _s(t) real part functions and imaginary part function, t representes the independent variable of multiple wavelet function, and s representes decomposition scale, and β representes zooming parameter, and the experience value is 0.83, to obtain near optimum filter effect.Can find out from formula (1); Real part

is the second derivative of Gaussian function, is used for the matching detection cut; And imaginary part is the first order derivative of Gaussian function; Be desirable Canny edge detection operator, can be used to locate the edge of cut.

303: According to the complex wavelet function calculates the integer pixel filter function

and half-integer pixel filter function

Wherein, the half-integer pixel be image x ∈ Z+1/2 (here with integer and 1/2 be defined as half-integer) pixel value located.Because in actual conditions, (m n) only is defined on the integer grid digital picture f, but the cut edge then possibly be between the pixel, so the Ridgelet conversion has half-pixel accuracy.That is, need to obtain image ridgelet conversion coefficient r _s(x), in the value of x ∈ Z and x ∈ Z+1/2 (half-integer), wherein Z is a set of integers.Can realize that it defines as follows through the filter function

of integer pixel point and the filter function

of half-integer pixel:

$\left\{\begin{array}{c}{H}_s^{\left(I\right)}\left(n\right)=\&Integral;\ker \left(t\right)\&CenterDot;{\mathrm{\&psi;}}_s(n-t)\mathrm{dt}\\ H_s^{\left(H\right)}\left(n\right)=\&Integral;\ker \left(t\right)\&CenterDot;{\mathrm{\&psi;}}_s(n+1/2-t)\mathrm{dt}\end{array}\right.---\left(2\right)$

In the formula, t is an argument of function, and the horizontal coordinate of n presentation video (discrete), value are integer, s=s ₁..., s _JBe a series of decomposition scale parameters, its experience width range (it is generally acknowledged 1～5 pixel) decision by cut; Complex wavelet function ψ _sImplication and the ψ in the formula (1) _sIdentical, ker (t) is a convolution kernel, and the embodiment of the invention is got ker (t)=1 _[1/2,1/2]

304: the filter function

of each yardstick and the projection P f (n) of

and vertical direction are made convolution, obtain the value of the monochromatic multiple Ridgelet coefficient

of each yardstick at half-pix point.

Wherein, The monochromatic multiple Ridgelet coefficient

of each yardstick is

value at x ∈ Z and x ∈ Z+1/2 in the value of half-pix point, and wherein Z is a set of integers.

Wherein, obtain

through following formula (3)

$r_s^a\left(x\right)=\left\{\begin{array}{c}{H}_s^{\left(I\right)}\left(n\right)*\mathrm{Pf}\left(n\right);x\&Element;Z\\ H_s^{\left(H\right)}\left(n\right)*\mathrm{Pf}\left(n\right);x\&Element;Z+1/2\end{array}\right.---\left(3\right)$

In the formula (3), * is a discrete convolution;

is monochromatic multiple ridgelet conversion coefficient; X is an independent variable; Its value is x ∈ Z and x ∈ Z+1/2; N is the discrete levels coordinate, and value is an integer;

defined image at the locational coefficient value of integral point and image at the locational coefficient value of half-integer point (x ∈ Z+1/2).

Through said method, get access to monochromatic Ridgelet coefficient

R, G, B three primary colours again and have corresponding

respectively

305: according to the multiple Ridgelet coefficient of monochrome

The multiple Ridgelet coefficient r of the fusion of computed image _s(x).

Wherein, obtaining the multiple Ridgelet coefficient of the corresponding respectively monochrome of R, G, B three primary colours

After, real part Re [] and the imaginary part Im [] to coefficient handles respectively according to following mode, obtains the fusion coefficients r of image _s(x):

$\left\{\begin{array}{c}\mathrm{Re}\left[r_s\left(x\right)\right]=\sqrt{{\left(\mathrm{Re}\right[r_s^{\left(R\right)}\left(x\right)\left]\right)}^2+{\left(\mathrm{Re}\right[r_s^{\left(G\right)}\left(x\right)\left]\right)}^2+{\left(\mathrm{Re}\right[r_s^{\left(B\right)}\left(x\right)\left]\right)}^2}\\ \mathrm{Im}\left[r_s\left(x\right)\right]=\sqrt{{\left(\mathrm{Im}\right[r_s^{\left(R\right)}\left(x\right)\left]\right)}^2+{\left(\mathrm{Im}\right[r_s^{\left(G\right)}\left(x\right)\left]\right)}^2+{\left(\mathrm{Im}\right[r_s^{\left(B\right)}\left(x\right)\left]\right)}^2}\end{array}\right.,s=s_1,...,s_J---\left(4\right)$

Wherein, s=s ₁..., s _JBe a series of decomposition scale parameters, x representes the independent variable in the conversion coefficient, and its value is x ∈ Z and x ∈ Z+1/2, and Z is a set of integers.Through type (4) merges the multiple Ridgelet coefficient of three single color component half-pixel accuracy, obtains the multiple Ridgelet coefficient of fusion of image.

306: according to fusion coefficients r _s(x) imaginary part extreme point is searched the interval at following possible cut of s yardstick place.

Wherein, the imaginary part of multiple wavelet function

Be Canny edge detection operator (seeing formula 1), with fusion coefficients r _s(x) imaginary part (Im [r _s(x)]) extreme point is designated as x _{S, i}, 1≤i≤K _sx _{S, i}I extreme point under the expression s yardstick, K _sBe all extreme point numbers under the s yardstick; Because imaginary part extreme point x _{S, i}The corresponding vertically position and the intensity of Gray Level Jump, therefore, being defined in following of s yardstick might vertical straight line (comprising vertical cut and other vertical edge, for example edge of image) be L in the interval of horizontal coordinate _{S, i}: [x _{S, i}, x _{S, i+1}], 1≤i＜K _sNeed to prove, under a s yardstick, coefficient r _s(x) possibly there are a lot of extreme points in all imaginary parts.And arbitrarily a pair of adjacent extreme value (being minimal value and maximum value) all corresponding one possibly be the interval of cut, and cut of an interval delineation of cut.

Above-mentioned need to prove, the edge of the interval of cut, cut and cut is the notion that is mutually related, the edge delineation cut of cut is interval, and cut is positioned at the cut interval.

307: might cut the interval in according to the pre-conditioned interval of confirming remarkable straight line place.

Wherein, present embodiment according to below pre-conditioned but be not limited to that these are pre-conditioned, confirm the interval at vertical cut place, in this process, vertical cut is defined as remarkable straight line.Then confirm remarkable straight line place interval as follows:

(1) significantly there is and only has a r in the interval at straight line place _s(x) mould maximal point e _i∈ [x _{S, i}, x _{S, i+1}], and | r _s(e _i) |>=T _Ci, wherein, | r _s(e _i) | expression r _s(x) mould maximal point; T _CiFor to interval L _{S, i}The distribution threshold value that the data set that all mould maximal point data are formed near the neighborhood is selected can rule of thumb artificial setting.

(2) remarkable two end points x in the interval at straight line place _{S, i}, x _{S, i+1}Satisfy Im [r _s(x _{S, i})]>=T _EiAnd Im [r _s(x _{S, i+1}]>=T _Ei, T wherein _EiFor to interval L _{S, i}The distribution threshold value that the data set that all imaginary part maximal point data are formed near the neighborhood is selected can rule of thumb artificial setting.

(3) | x _{S, i}-x _{S, i+1}|≤W _Max, W _MaxIt is the breadth extreme of cut.

(4) there is and only has a real part Re [r in the interval _s(x)] maximal point m _i∈ [x _i, x _I+1], and it satisfies | Re [r _s(m _i)]/Im [r _s(m _i)] |>=T _θ, T wherein _θFor to interval L _{S, i}The distribution threshold value that the data set that all real part maximal point data are formed near the neighborhood is selected can rule of thumb artificial setting.

In the embodiment of the invention, significantly straight line is a cut, describe for ease and understand, with the section definition of confirming the cut place for confirming the interval at remarkable straight line place, the two expression equivalent in meaning.

Comprehensively above-mentioned; Significantly the principle of straight line (cut) detection comprises: in plural Ridgelet function, real part

is mainly used in the matching detection of vertical cut; Imaginary part

then is mainly used in the location at vertical cut edge.Compare with real number field Ridgelet detection algorithm of the prior art, the complex field method can provide mould value and phase information.In fact, because the Gray Level Jump that mould value that provides in the complex field Ridgelet detection algorithm and cut itself cause is approaching, so each mould value | r _s(x) | maximal point is corresponding to cut, and the intensity of the big or small corresponding cut of mould value, and the mould value is big more, and the expression cut is obvious more.The corresponding vertically Gray Level Jump of the imaginary part maximal point of Ridgelet coefficient then can effectively be distinguished vertical cut and vertical edge in the vertical Gray Level Jump through phase information.Particularly, each mould value maximal point can be unique corresponding to a cut, and because to r _s(x) no matter delivery is the light and shade cut, all can be corresponding to the maximal point of a mould, thus bright, dark cut is detected simultaneously; Under the ideal situation, a vertically pulse on the corresponding horizontal projection of cut is so corresponding mould maximal point phase place is 0 °/180 ° (bright/dark cut); And vertically the edge corresponding step, so corresponding mould maximal point phase place is ± 90 °.

In the present embodiment, to image f (m, n) a ° rectilinear direction is done projection (the θ value is 0 in the present embodiment) along θ+90; Obtain 1 dimension projection signal; Be the multiple wavelet transformation of s to projection result as yardstick then, whole process can be called again that (m n) does plural Ridgelet conversion to image f.Because each point of 1 dimension projection signal is corresponding to vertically certain bar straight line in the image, so the real part of wavelet function can be used to detect cut in the Ridgelet conversion; Simultaneously, because the imaginary part of wavelet function is the Gaussian function first order derivative in the plural Ridgelet conversion, therefore it carry out above-mentioned mathematics manipulation with itself and one dimension projection signal corresponding to the Canny edge detection operator, can be used for the location at cut edge.

Simultaneously; The embodiment of the invention is obtained each single color component coefficient; Utilize each single color component information, the monochromatic component coefficient is merged the coefficient that the back obtains coloured image, and directly convert coloured image into gray scale image and compare; It can make full use of the color information of video, thereby improves detection efficiency.

Embodiment three

Present embodiment is on the basis of a last embodiment, at the multiple Ridgelet coefficient r of the later half integer precision of the fusion that obtains _s(x), x ∈ N ∪ (N+1/2) (s=s ₁..., s _JBe decomposition scale), detect behind the linear feature under each yardstick s, the remarkable straight line that a plurality of yardstick s are detected down merges, and confirms that final candidate's cut is interval.

As shown in Figure 4, the detection method process flow diagram of the image scratch that provides for present embodiment, this method comprises:

401: the interval that obtains the remarkable straight line place under multiple dimensioned.

Wherein, according to embodiment two, can obtain under the s yardstick might cut the interval in the remarkable interval at straight line place; When the value of decomposition scale s is s ₁..., s _JThe time, can obtain the interval at the remarkable straight line place under multiple dimensioned through the method for the foregoing description.Be specially:

Suppose Ridgelet coefficient r _s(x) at yardstick s _j(extract K altogether under 1≤j≤J) _jThe remarkable straight line of bar is designated as L _{J, i (k)}: [x _{J, i (k)}, x _{J, i (k)+1}], 1≤j≤J, 1≤k≤K _j, wherein, j is a variable, representes a certain yardstick; J representes total number of s yardstick; The interval sequence number of the remarkable straight line of i (k) expression k bar, K _jThe number of representing remarkable straight line.

402:, set up the crestal line of being formed by the imaginary part maximum point of multiple dimensioned Ridgelet coefficient according to the interval at remarkable straight line place.

Particularly, 402 can realize through following sequencing step:

With multiple dimensioned Ridgelet coefficient

All imaginary part maximum value are designated as OPEN (identifier is used for identifying the imaginary part maximum point that needs fusion); Wherein, s _j, 1≤j≤J is a series of decomposition scales;

As yardstick s _jFrom j=J to 1, and as remarkable straight line L _{J, i (k)}From k=1 to K _jThe time; Circulation is operated as follows, up to generating crestal line:

If remarkable straight line L _{J, i (k)}, left end point x _{J, i (k)}By mark CLOSE (identifier identifies Fused imaginary part maximum point), then Ridge number this point of process is designated as left _{J, i (k)}Otherwise with x _{J, i (k)}Beginning is set up one from s _jTo s ₁, the continuous Ridge that all forms by the imaginary part maximal point, this Ridge the imaginary part maximum point of process all be designated as CLOSE, charge to letf to this Ridge number _{J, i (k)}

In like manner, if L _{J, i (k)}Right endpoint x _{J, i (k)+1}By mark CLOSE, be designated as right to Ridge number that passes through this point _{J, i (k)}Otherwise with x _{J, i (k)+1}Beginning is set up one from s _jTo s ₁, the continuous Ridge that all forms by the imaginary part maximal point; This Ridge the imaginary part maximum point of process all be designated as CLOSE, charge to right to this Ridge number _{J, i (k)}

Through above-mentioned processing, L _{J, i (k)}The interval S of corresponding candidate's cut _{J, i (k)}, its left end point is that sequence number is left _{J, i (k)}Ridge in the position of the strongest imaginary part maximum point of amplitude; Right endpoint then is that sequence number is right _{J, i (k)}Ridge in the position of the strongest imaginary part maximum point of amplitude.

403: confirm the interval that cut belongs to according to crestal line, with the interval merging of all cuts.

Final candidate's scored area is the interval union of the corresponding cut of above-mentioned all remarkable straight lines, promptly satisfies $\mathrm{\&Omega;}=\left\{(m,n)\right|n\&Element;\underset{1\&le;j\&le;J,1\&le;k\&le;K_j}{\&cup;}{S}_{j,i\left(k\right)}\}$ Institute's pointed set.

The principle of above content is: the end points of each remarkable straight line is the imaginary part maximum point of conversion coefficient.According to the metric space filtering theory, when filter function adopted certain order derivative of Gaussian function, the very big/little value of its filtered can constitute a continuous crestal line (Ridge) from thick yardstick to thin yardstick in metric space.

Present embodiment is based on embodiment two, in 307 further, T _Ci, T _EiAnd T _θCan also obtain through the choosing method of distribution as unimodal threshold value." single mode " distributes, and is meant that the distribution statistics histogram of data has unimodal form.Embodiment of the invention Ridgelet conversion coefficient is concentrated, and most sample points are the small magnitude regional value that background produces, and only comprises a small amount of cut generation sample point and then has big amplitude, thereby need a threshold value that itself and background sample point are separated.For automatic selected threshold, need to make following hypothesis earlier: the background sample point is obeyed Weibull (Weibull) distribution W (r among the data set X; α, β), r＞0; And sample point (being produced by cut) is big amplitude " pollution " point away from this distribution.Weibull probability distribution function expression formula is P _{α, β}(r)=1-exp ((r/ α) ^β), r＞0, α wherein, β is respectively convergent-divergent, decay factor, and r is continuous independent variable, no Special Significance.

Because the p quantile of a probability distribution is the point that satisfies following character: promptly will be with Probability p less than this quantile corresponding to this distributed random variable.

Therefore, the automatic threshold of distribution as unimodal data set X chosen comprised:

The automatic threshold T of the sample point that background sample point and cut produce among the differentiation data set X _A(X) can be taken as Weibull distribution W (r; α, p quantile β):

T _A(X)＝α[ln(1/(1-p))] ^1/β(5)

Wherein, X representes to comprise the data set of the sample point that background sample point and cut produce, and α, β are respectively convergent-divergent, decay factor, and p is the quantile of Weibull distribution.

When through type (5) has been confirmed automatic threshold T _A(X) after the expression formula, under the situation that contains the singular point sample, sane estimation (sane estimation is promptly carried out parameter estimation under the situation that comprises the interference of noise and unusual sample point) parameter alpha, β:

Note M (X) be the median of X, and " smaller part generally interval " that D (X) distributes for X (promptly having a least bit to drop on weak point interval wherein).Then can try to achieve and M (X) the Weibull model parameter alpha that these two robust statistics of D (X) are the most consistent through the system of equations in the formula of separating (6); β; Be different Weibull distribution parameters, can be corresponding to different M (X), D (X); So the M (X) that utilizes statistics to draw, D (X) estimates only distribution parameter.Formula (6) is:

$\left\{\begin{array}{c}\frac{d}{\mathrm{dr}}[P_{\mathrm{\&alpha;},\mathrm{\&beta;}}^{-1}(1/2+P_{\mathrm{\&alpha;},\mathrm{\&beta;}}\left(r\right))-r]=0\\ P_{\mathrm{\&alpha;},\mathrm{\&beta;}}^{-1}(1/2+P_{\mathrm{\&alpha;},\mathrm{\&beta;}}\left(r\right))-r=D\left(X\right)\\ P_{\mathrm{\&alpha;},\mathrm{\&beta;}}^{-1}(1/2)=M\left(X\right)\end{array}\right.---\left(6\right)$

Wherein, P _{α, β}(r)=1-exp ((r/ α) ^β) be the Weibull probability distribution function.Formula (6) can solve through the process of iteration of nonlinear equation, and initial solution can be established W (r; α; β) obtain for Rayleigh distribution (β=2); (α at this moment; β r) is

owing to there be " points of contamination " of big amplitude in the data, so traditional maximum-likelihood method and inapplicable; And this method through finding the solution the system of equations based on robust statistic, can estimate distribution parameter accurately.

To sum up, the selection of threshold step to distribution as unimodal data set X is:

1),, steadily and surely estimates Weibull distribution W (r to noisy distribution as unimodal data set X according to formula (6); α, model parameter α β), β.

2), select the threshold value of the p quantile of this model as data set X according to formula (5).

T _Ci, T _EiAnd T _θCan obtain through the choosing method of above-mentioned distribution as unimodal threshold value.

In the present embodiment, behind the interval that obtains remarkable straight line place, set up the crestal line of being made up of the imaginary part maximum point of multiple dimensioned Ridgelet coefficient, each point has been represented detected cut frontier point under corresponding scale on the crestal line.And amplitude point of maximum intensity has wherein been represented the strongest skirt response (i.e. the optimum detection yardstick at this edge), thereby can be used as the accurate estimation at cut edge; Simultaneously; Detect through the cut of said method, and selected cut is interval, can adapts to the cut of different in width different scale.

In addition, adopt the choosing method of distribution as unimodal threshold value to obtain T in the present embodiment _Ci, T _EiAnd T _θ, can avoid the artificial blindness that rule of thumb is worth setting threshold, help improving the robustness of scratch detection.

The method that provides through present embodiment; Image is done plural Ridgelet conversion obtain multiple Ridgelet coefficient; Wherein, the position and the intensity of the Gray Level Jump of the imaginary part extreme point correspondence image of multiple Ridgelet coefficient can be used for the cut edge and the vertical location of image border; And the phase information of the multiple Ridgelet coefficient that combination imaginary part coefficient and real part coefficient obtain then can be used to distinguish the edge of vertical cut and vertical step, thereby finally can improve the degree of accuracy of location cut; The different gray-scale values of the mould value correspondence image of the multiple Ridgelet coefficient that obtains in conjunction with imaginary part coefficient and real part coefficient can detect cut.Simultaneously; After the embodiment of the invention obtains each single color component coefficient; Utilize each single color component information, the monochromatic component coefficient is merged the coefficient that the back obtains coloured image, and directly convert coloured image into gray scale image and compare; Can make full use of the color information of video, thereby improve detection efficiency.

Embodiment four

Present embodiment will utilize the field colouring information, remove the false cut that possibly exist among the interval Ω of candidate's cut among the last embodiment.(m n) demarcates cut, and bianry image is to have only 0,1 two value in the image, and on duty is the true cut point of 1 interval scale can to define a bianry image SM unidimensional with image.

To any point r in the image ₀=(m ₀, n ₀), wherein, m ₀Be vertical coordinate, n ₀Be horizontal coordinate; Define its neighborhood color distance as follows: at m ₀-1, m ₀, m ₀The non-scored area of+1 these 3 row (among the SM (m, n)=0), is got apart from r ₀6 nearest pixels, and (m ₀, n ₀-1), (m ₀, n ₀+ 1) the non-cut point in constitutes r ₀Neighborhood point set N (r ₀).The neighborhood color distance is defined as:

Wherein || || ₂Be the second order norm, f (r _i) be pixel r _iThe coloured image value, this value is for vector.

Based on foregoing, as shown in Figure 5, the method flow diagram of the false cut of the eliminating that provides for present embodiment; In order to improve the accuracy rate of detection; This method will be utilized colouring information, and the false cut point among the interval Ω of the candidate's cut that obtains among the embodiment three is removed, and this method comprises:

501: (m n) is initialized as the point among all candidate's scored area Ω is designated as true cut point with SM.

502: the neighborhood color distance threshold value T that obtains the point set of all NCD formations _s

Wherein, each pixel all has a NCD (neighborhood color distance), T _sCan rule of thumb be worth setting, also can select threshold method to obtain through the distribution as unimodal among the last embodiment, particularly, through type (5) and formula (6) be obtained.

503: according to neighborhood color distance threshold value T _sGet rid of false cut point.

Particularly, can realize through following steps:

When still having true cut point, promptly SM (m, some n)=1 be changed to non-cut point SM (m, n)=0; Following two steps operation is carried out in circulation, up to getting rid of false cut point:

1) each is satisfied SM (m ₀, n ₀The point r of)=1 ₀, if NCD (r ₀)＜T _s, then put SM (m ₀, n ₀)=0 is to remove false cut point;

2) to n from 1 to Z (integer), if (∑ _1≤m≤MSM (m, n))/M＜T _SD, then put SM (m, n)=0,1≤m≤M, to remove false cut.Wherein, the columns of M presentation video; Threshold value T _SDFor minimum allows cut density, can rule of thumb be worth setting.

Through said method, (m, some n)=1 is the cut point finally can to confirm SM.

In the present embodiment, use bianry image information and distribution as unimodal to select threshold method to combine, get rid of the false cut in the bianry image, realize verification, reduced false alarm rate testing result.

Embodiment five

The method that present embodiment provides detects the cut with small vertical inclination angle.Suppose that cut is θ apart from the vertical direction inclination maximum _Max, then at H=1/tan θ _MaxLocal vertically scope in, the horizontal-shift of cut is no more than a pixel, so can think that it is vertical fully.General θ _Max＜5 °, H gets 16.Therefore, can utilize the method for the foregoing description, image is grown the local vertically scratch detection for H.Concrete grammar still can be divided into: single color component half-pixel accuracy optimum complex Ridgelet filtering of adopting embodiment two to provide, and each color component ridglet coefficient merges; The method that adopts embodiment three to provide is carried out the remarkable straight-line detection of single scale, and multiple dimensioned testing result merges; The method that can also adopt embodiment four to provide is carried out the testing result verification.Below only to narrating with the foregoing description difference.

As shown in Figure 6, the detection method process flow diagram of the image scratch that provides for present embodiment, this method comprises:

601: (m n) is divided into a series of high H of being, and overlapped subgraph with coloured image f.

Wherein, (m, n) mode according to formula (7) is divided into a series of high be H, the subgraph f of overlapped H/r with coloured image f ^(j)(m, n), formula (7) as follows:

f ^(j)(m，n)＝f(m+j·H/r，n)；0≤m＜H，

Wherein, (m n) is pixel (m, coloured image value n) to the implication of each parameter: f; H is the subgraph height; H/r is the overlapping height of subgraph; M is that the height of entire image (is the line number of image; Height with the line number uncalibrated image), j is the subgraph sequence number,

represent Mr/H is rounded up.

Overlapping subgraph piecemeal synoptic diagram after cutting apart is as shown in Figure 7.

602: each subgraph is carried out image scratch detect, obtain the candidate scored area of image in the vertical scope of subgraph.

Wherein, to each subgraph f ^(j)(m n) carries out image scratch and detects, and obtains image f (m, n) the candidate's scored area S in [jH/r, jH/r+H-1] vertical scope _{I, j}, 1≤i≤N ^(j),

The columns of N presentation video.

Particularly, to each subgraph f ^(j)(m; N) carry out image scratch according to the mode of previous embodiment and detect, comprise each subgraph is carried out single color component half-pixel accuracy optimum complex Ridgelet filtering, each color component ridglet coefficient merges; The remarkable straight-line detection of single scale, multiple dimensioned testing result merges; Through aforesaid operations, obtain image f (m, n) the candidate's scored area S in [jH/r, jH/r+H-1] vertical scope _{I, j}, 1≤i≤N ^(j),

603: all scored area are merged, obtain candidate's scored area Ω of entire image.

604: candidate's scored area is carried out verification as a result.

Wherein, candidate's scored area Ω of the entire image that obtains is carried out the verification of testing result, the method for its verification and embodiment four described methods are basic identical; Difference is; Among the relative embodiment four 503, the concrete grammar of getting rid of false cut is that (m n) does [90 °-θ to SM _Max, 90 °+θ _Max] the Hough conversion of angular range, wherein, θ _MaxFor cut leaves the vertical direction inclination maximum; (m, n) middle semi-invariant surpasses threshold value T SM _SD(T _SDFor minimum allows cut density, can rule of thumb be worth settings) straight line the point of process compose 1, all the other some taxes 0, then (m is that 1 point is true cut point n) to SM.Wherein, the Hough conversion is to ask bianry image (0 or 1) accumulated value of straight line under certain angle, if SM (m n) has TSD cut point in certain rectilinear direction, then is judged to true cut, and the cut point that is judged to a little on this straight line.

In the present embodiment; The image segmentation that will have cut with angle becomes a series of subgraphs, because therefore the cut near normal in each subgraph carries out scratch detection to each subgraph according to aforesaid method; Scored area with the candidate merges afterwards, obtains candidate's scored area of entire image.Therefore, through the method for present embodiment, can detect the cut that has small vertical inclination angle.

Need to prove that the embodiment of the invention also can be used for the scratch detection of black-and-white film image, because the black-and-white film image has only brightness, is scalar, therefore need not the fusion of a color component Ridgelet coefficient.

Embodiment six

As shown in Figure 8, the pick-up unit structural drawing of the image scratch that provides for present embodiment, this device comprises:

Projection module 81 is done projection with the single color component in the two dimensional image along the cut direction, obtain the projection of one dimension cut with;

Computing module 82 is according to the filter function of multiple wavelet function calculating pixel point;

Processing module 83, with filter function and the projection of one dimension cut with make convolution, obtain the multiple ridge wave system number of single color, count the multiple ridge wave system number of fusion of computed image according to the multiple ridge wave system of single color;

Operational module 84 is confirmed the edge of cut according to merging again the imaginary part of ridge wave system number.Wherein, projection module 81 obtain the projections of one dimension cut with afterwards, obtain the multiple wavelet function that is used to detect cut according to pre-set criteria; Computing module 82 is according to the filter function that is somebody's turn to do multiple wavelet function calculating pixel point, and this pixel comprises integer pixel point and half-integer pixel; Processing module 83 with filter function and the projection of one dimension cut with make convolution, obtain the multiple ridge wave system number of single color, through the formula in the embodiment of the invention two (4), count the multiple ridge wave system number of fusion of computed image according to the multiple ridge wave system of single color; Operational module 84 is confirmed the edge of cut according to the imaginary part that merges multiple ridge wave system number, wherein, searches the interval that institute might the cut place according to the imaginary part extreme point of fusion coefficients, confirms the interval that cut belongs to through preset condition again.Concrete grammar sees embodiment two for details, repeats no more here.

As shown in Figure 9, further, this device can also comprise:

Acquisition module 91 obtains the multiple wavelet function that is used to detect cut.

Further, this device can also comprise:

Multiple dimensioned processing module 92 is used to obtain the interval that the cut under a plurality of decomposition scales belongs to; According to the interval at cut place, set up the crestal line of forming by the imaginary part maximum point of the multiple ridge wave system number under a plurality of decomposition scales; Merge according to the interval of crestal line the place of the cut under a plurality of decomposition scales.

Wherein, the handling principle of multiple dimensioned processing module 92 comprises: the end points of each remarkable straight line (the vertical cut of acute pyogenic infection of finger tip in embodiments of the present invention) is the imaginary part maximum point of conversion coefficient.According to the metric space filtering theory, when filter function adopted certain order derivative of Gaussian function, the very big/little value of its filtered can constitute a continuous crestal line (Ridge) from thick yardstick to thin yardstick in metric space.

Preferably, this device can also comprise:

Verification module 93 is used to obtain the neighborhood color distance threshold value of the point set that the neighborhood color distance constitutes, and the neighborhood color distance is a pixel neighborhood of a point color distance in the image; Get rid of false cut point according to neighborhood color distance threshold value.

Wherein, verification module 93 is utilized the field colouring information, can remove the false cut that possibly exist among the interval Ω of cut.(m n) demarcates cut, and bianry image is to have only 0,1 two value in the image, and on duty is the true cut point of 1 interval scale can to define a bianry image SM unidimensional with image.Through definition pixel neighborhood of a point color distance, and obtain the neighborhood color distance threshold value of the point set that the neighborhood color distance constitutes, get rid of false cut point according to neighborhood color distance threshold value.Wherein, neighborhood color distance threshold value can select threshold method to obtain according to the distribution as unimodal in the embodiment of the invention, also can rule of thumb be worth setting.

When cut has small vertical inclination angle, in order to detect cut, can utilize above-mentioned device with tiny inclination angle, image is grown the local vertically scratch detection for H, concrete method sees method embodiment for details, repeats no more here.

Therefore, on the basis of Fig. 8, this device also comprises: cut apart module 11 and merge module 12, shown in figure 10:

Cut apart module 11, two dimensional image is divided into overlapped subgraph;

Wherein, color images is become a series of high H of being, and overlapped subgraph;

Afterwards, each subgraph is carried out image scratch according to the method for the embodiment of the invention two detect, obtain the candidate scored area of image in the vertical scope of subgraph; Scored area with all subgraphs merges again, and therefore, correspondingly, device also comprises:

Merge module 12, be used for the cut of all subgraphs is merged, obtain the cut of image.

In the present embodiment, image is done plural Ridgelet conversion along rectilinear direction, because each point of 1 dimension projection signal is corresponding to vertically certain bar straight line in the image, so the real part of wavelet function can be used to detect cut in the Ridgelet conversion; Simultaneously; Because the imaginary part of wavelet function is the Gaussian function first order derivative in the plural Ridgelet conversion; It is corresponding to the Canny edge detection operator; Therefore itself and one dimension projection signal are carried out above-mentioned mathematics manipulation, can be used for the location at cut edge, thereby finally can improve the degree of accuracy of location cut.Simultaneously, present embodiment can make full use of the color information of image through the fusion coefficients of monochromatic component coefficient acquisition coloured image, improves detection efficiency.

It will be appreciated by those skilled in the art that accompanying drawing is the synoptic diagram of an embodiment, module in the accompanying drawing or flow process might not be that embodiment of the present invention is necessary.

It will be appreciated by those skilled in the art that the module in the device among the embodiment can be distributed in the device of embodiment according to the embodiment description, also can carry out respective change and be arranged in the one or more devices that are different from present embodiment.The module of the foregoing description can be merged into a module, also can further split into a plurality of submodules.

The invention described above embodiment sequence number is not represented the quality of embodiment just to description.

Part steps in the embodiment of the invention can utilize software to realize that corresponding software programs can be stored in the storage medium that can read, like CD or hard disk etc.

The embodiment of the invention can realize that corresponding software can be stored in the storage medium that can read, for example in the hard disk of computing machine, CD or the floppy disk through software.

The above is merely preferred embodiment of the present invention, and is in order to restriction the present invention, not all within spirit of the present invention and principle, any modification of being done, is equal to replacement, improvement etc., all should be included within protection scope of the present invention.

Claims (12)

1. the detection method of an image scratch is characterized in that, said method comprises:

Single color component in the two dimensional image is done projection along the cut direction, obtain the projection of one dimension cut with;

Filter function according to multiple wavelet function calculating pixel point; The real part of said multiple wavelet function adopts the second derivative form of Gaussian function; Be used for the matching detection cut, the imaginary part of said multiple wavelet function is the first order derivative of Gaussian function, is used to locate the edge of cut;

With said filter function and the projection of said one dimension cut with make convolution, obtain the multiple ridge wave system number of single color, count the multiple ridge wave system number of fusion of computed image according to the multiple ridge wave system of said single color;

Confirm the edge of cut according to the imaginary part of the multiple ridge wave system number of said fusion.

2. method according to claim 1 is characterized in that, the step of the filter function of the multiple wavelet function calculating pixel point of said basis comprises:

According to the filter function of said multiple wavelet function computes integer pixel and the filter function of half-integer pixel.

3. method according to claim 1 is characterized in that, said imaginary part according to the multiple ridge wave system number of said fusion confirms that the step at the edge of cut comprises:

According to the imaginary part extreme point that merges multiple ridge wave system number obtain cut the interval that might belong to;

Interval according to pre-conditioned definite cut place.

4. method according to claim 3 is characterized in that, the step in said interval according to pre-conditioned definite cut place comprises:

If a mould maximal point that merges multiple ridge wave system number is arranged in the interval at cut place, and the value of said mould maximal point is more than or equal to the distribution threshold value of the data set of all mould maximal point data compositions near the field said interval;

If the imaginary part extreme value of two end points in the interval at cut place is more than or equal to the distribution threshold value of the data set of all imaginary part maximal point data compositions near the field said interval;

If the difference between two end points in the interval at cut place is smaller or equal to the breadth extreme of cut; With

If there is a maximal point that merges the real part of multiple ridge wave system number in the interval at cut place; And the real part of said point and the ratio of extreme values of imaginary part confirm then that more than or equal to the distribution threshold value of the data set of all real part maximal point data compositions near the field said interval said cut is in described interval.

5. method according to claim 4 is characterized in that, the distribution threshold value of said data set comprises:

Make up the Weibull probability distribution function of data set, estimate the convergent-divergent and the decay factor of Weibull Function;

Obtain the distribution threshold value of the quantile of Weibull probability distribution function according to said convergent-divergent and decay factor as data set.

6. according to claim 3 or 4 described methods, it is characterized in that said method also comprises:

Obtain the interval at the cut place under a plurality of decomposition scales;

According to the interval at said cut place, set up the crestal line of forming by the imaginary part maximum point of the multiple ridge wave system number under said a plurality of decomposition scales;

Merge according to the interval of said crestal line the place of the cut under a plurality of decomposition scales.

7. method according to claim 6 is characterized in that, said method also comprises:

Obtain the neighborhood color distance threshold value of the point set of neighborhood color distance formation, said neighborhood color distance is a pixel neighborhood of a point color distance in the image;

Get rid of false cut point according to said neighborhood color distance threshold value.

8. method according to claim 1 is characterized in that, said single color component in the two dimensional image is done also to comprise before the projection along the cut direction:

Two dimensional image is divided into overlapped subgraph;

Correspondingly, said imaginary part according to the multiple ridge wave system number of said fusion confirms that the edge of cut also comprises afterwards:

All subgraphs are merged, obtain the cut of image.

9. the pick-up unit of an image scratch is characterized in that, said device comprises:

Projection module is done projection with the single color component in the two dimensional image along the cut direction, obtain the projection of one dimension cut with;

Computing module; According to the filter function of multiple wavelet function calculating pixel point, the real part of said multiple wavelet function adopts the second derivative form of Gaussian function, is used for the matching detection cut; The imaginary part of said multiple wavelet function is the first order derivative of Gaussian function, is used to locate the edge of cut;

Processing module, with said filter function and the projection of said one dimension cut with make convolution, obtain the multiple ridge wave system number of single color, count the multiple ridge wave system number of fusion of computed image according to the multiple ridge wave system of said single color;

Operational module is confirmed the edge of cut according to the imaginary part of the multiple ridge wave system number of said fusion.

10. device according to claim 9 is characterized in that, said device also comprises:

Multiple dimensioned processing module is used to obtain the interval that the cut under a plurality of decomposition scales belongs to;

According to the interval at said cut place, set up the crestal line of forming by the imaginary part maximum point of the multiple ridge wave system number under said a plurality of decomposition scales;

Merge according to the interval of said crestal line the place of the cut under a plurality of decomposition scales.

11. device according to claim 9 is characterized in that, said device also comprises:

The verification module is used to obtain the neighborhood color distance threshold value of the point set that the neighborhood color distance constitutes, and said neighborhood color distance is a pixel neighborhood of a point color distance in the image;

Get rid of false cut point according to said neighborhood color distance threshold value.

12. device according to claim 9 is characterized in that, said device also comprises:

Cut apart module, two dimensional image is divided into overlapped subgraph;

Correspondingly, said device also comprises:

Merge module, be used for confirming that according to the imaginary part of the multiple ridge wave system number of said fusion the edge of cut merges all subgraphs afterwards, obtains the cut of image.

Images