CN101866484A - Method for computing significance degree of pixels in image - Google Patents

Abstract

The invention discloses a method for computing the significance degree of pixels in an image. Aiming at each pixel in the image, a group of adjacent areas in a multi-scale range are designed; by adopting the L*a*b* color characteristics, pixels, which have characteristic difference from the pixel, in each adjacent area are respectively computed; and the number proportion of the pixels with the difference in each adjacent area is taken as the significance degree of the pixel. The multi-scale computation of the pixels is taken into consideration, and the method has the advantages of simple computation and easy implementation. The obtained image significance degree distribution map has the same resolution with the original image, and has more reasonability in visual sense, so the input with good robustness is provided for the significance degree-based computational vision application.

Description

The computing method of significance degree of pixels in a kind of image

Technical field

The invention belongs to computer vision field, be specifically related to a kind of computing method significance degree of pixels in the image.

Background technology

And cognitive science studies show that at heart, when people observe piece image, our vision system can be primarily focused on one or several significant zone apace, and then further observes the content in the image, and these significant zones are our interesting areas often.At present, the computation vision field that is extracted in to marking area in the image has a wide range of applications, for example, extract (Content-based Image Retrieval at content-based image, CBIR) in the research, if what images match adopted is the feature of extracting from marking area, Pi Pei effect and final extraction hit rate are better than the feature that employing is extracted from entire image so.Other application also comprises, adapt in vain Image Compression (as, JPEG2000), the white adaptation browser technology of video, Target Recognition adapts to image zoom technology etc. in vain.In fact, these technology have a common hypothesis, promptly, the main information content in the image has all concentrated in the marking area of image, the feature of extracting from the marking area of image not only can be for application provides enough information, the interfere information that can also avoid those to be brought by non-marking area.Therefore, the extraction to marking area in the image has important researching value.

At present, in the computation vision field, in order to extract the marking area in the image, usually the method that adopts is, by the significance degree of each pixel in the computed image, obtain the significance degree distribution plan (saliency map) of image earlier, then this remarkable distribution plan is got the computing of threshold value, obtain the image segmentation result of a two-value, foreground area wherein is the marking area that will extract.In said process, the effect important influence that the significance degree distribution plan extracts final marking area.At present, about the calculating of significance degree of pixels in the image, mainly be divided into based on the method for physiological mode with based on the method for computation model, usually, the significance degree distribution plan resolution that is obtained by these methods is lower.Recently, A Kante Achant (reference: R.Achanta, S.Hemami, F.Estrada, et al.Frequency-tunedsalient region detection.IEEE Conf.on CVPR, 2009) simple computing method have been proposed, though the significance degree distribution plan that is obtained by this method has reached the resolution identical with original image, but in some cases, the visually-perceptible from the people seems that some is unreasonable.

Summary of the invention

The object of the present invention is to provide the significance degree computing method of pixel in a kind of image, the image significance degree distribution plan that is calculated by this method not only has the resolution identical with original image, on visually-perceptible, also have more simultaneously rationality, thereby provide robustness good input for using based on the computation vision of significance degree.

The significance degree computing method of pixel in a kind of image, for each pixel p in the image, calculate as follows:

(1) asks for neighborhood zone in one group of multiple dimensioned scope of pixel p; Wherein,

The neighborhood zone nearest apart from pixel p is a border circular areas R ₀(p), be defined as follows:

R ₀(p)＝{q|0≤‖p-q‖ ₂≤r ₀，q∈∧}

‖ ‖ ₂Expression Euclidean distance tolerance, all pixel in the ∧ presentation video, r ₀Radius for this border circular areas;

Border circular areas R ₀(p) it is the annular region that concentric circles surrounded in the center of circle with the pixel p that the neighborhood zone outside is k, and they are defined as follows:

R _i(p)＝{t|r _i-1≤‖p-t‖ ₂≤r _i，t∈∧}，i＝1，...，k

r _iIt is ring-like region R _i(p) outer boundary is to the radius of p,

W is the width of image, and H is the height of image, and the interregional radius step-length difference of adjacent annular is constant;

(2) utilize the L of pixel in the image ^*a ^*b ^*Color characteristic calculates in above-mentioned each neighborhood zone respectively and pixel p has the pixel of difference degree;

(2.1) at border circular areas R ₀(p) in, has the pixel D of difference degree with pixel p ₀(p) be characterized by:

I wherein _p,

Remarked pixel p,

L ^*a ^*b ^*Color feature value,

Wherein

With

Be respectively in the image all pixels at L ^*a ^*b ^*Variance in the color space on three passages;

(2.2) at each annular region R _i(p) in, has the pixel D of difference degree with pixel p _i(p) be characterized by:

Wherein

Remarked pixel

L ^*a ^*b ^*Color feature value,

Or, || _CardThe gesture of expression set, I _{Q '}The L of remarked pixel q ' ^*a ^*b ^*Color feature value;

(3) significance degree of calculating pixel p $S\left(p\right)=\frac{1}{k}\underset{i=1}{\overset{k}{\mathrm{\Σ}}}\frac{{\left|D_i\left(p\right)\right|}_{\mathrm{card}}}{{\left|R_i\left(p\right)\right|}_{\mathrm{card}}}.$

Technique effect of the present invention is embodied in: from the definition of S (p) as can be seen, the present invention about the significance degree of pixel p calculate be exactly in k annular region of pixel p with the mean value of the dissimilar number of pixels proportion of pixel p, this definition not only simply but also easily realized, S (p) can be regarded as a wave filter, then image is carried out filtering and get final product, when realizing, can adopt parallel processing method to quicken calculating process.Resulting image significance degree distribution plan not only has identical resolution with original image, and visually has more rationality, thereby provides robustness good input for using based on the computation vision of significance degree.

Description of drawings

Fig. 1 is the inventive method process flow diagram;

Fig. 2 is the neighborhood area schematic in one group of multiple dimensioned scope of the present invention;

Fig. 3 is the effect contrast figure who adopts distinct methods to calculate to six examples, wherein Fig. 3 (a) is six secondary real natural images, Fig. 3 (b) is for adopting document (L.Itti, C.Koch, E.Niebur.A model of saliency based visual attention for rapid scene analysis.IEEETrans.on PAMI., 1998,20 (11): the 1254-1259) result that obtains of method, Fig. 3 (c) is for adopting document (X.Hou, L.Zhang.Saliency detection:A spectral residualapproach.IEEE Conf.on CVPR, 2007) result that obtains of method, Fig. 3 (d) is for adopting document (R.Achanta, S.Hemami, F.Estrada, et al.Frequency-tunedsalient region detection.IEEE Conf.on CVPR, 2009) result that obtains of method, the result of Fig. 3 (e) for adopting the inventive method to obtain.

Embodiment

The present invention is further detailed explanation below in conjunction with accompanying drawing.

As shown in Figure 1, detailed process is:

(1) a given width of cloth natural image I for each pixel p in this image, at first tries to achieve the neighborhood zone in one group of multiple dimensioned scope of this pixel p.Wherein nearest apart from pixel p neighborhood zone is a border circular areas R ₀(p), as shown in Figure 2, it is defined as follows:

R ₀(p)＝{q|0≤‖p-q‖ ₂≤r ₀，q∈∧}

‖ ‖ wherein ₂Expression Euclidean distance tolerance, all pixel in the ∧ presentation video, r ₀Radius for this border circular areas.

Except above-mentioned border circular areas R ₀(p) outside, it is the k that concentric circles a surrounded annular region in the center of circle with the pixel p that remaining neighborhood zone is one group, and as shown in Figure 2, they are defined as follows:

R _i(p)＝{t|r _i-1≤‖p-t‖ ₂≤r _i，t∈∧}，i＝1，...，k

R wherein _iIt is ring-like region R _i(p) outer boundary is to the radius of p, and among the present invention, for this k annular region, we make r _i=r _I-1+ Δ r, and require maximum radius r _kSatisfy

Wherein W is the width of image I, and H is the height of image I.For this group relation, if we have set annular region number k and radius r ₀Value, the value of Δ r is easy to obtain so.

(2) utilize the L of pixel in the image ^*a ^*b ^*Color characteristic calculates in above-mentioned each neighborhood zone and pixel p has all pixels of certain difference degree.

(2.1) wherein at border circular areas R ₀(p) in, for having those pixels D of certain difference degree with pixel p ₀(p), we define its computing formula and are:

I wherein _p,

Remarked pixel p,

L ^*a ^*b ^*Color feature value, ‖ ‖ ₂Refer to Euclidean distance tolerance, and

Wherein

With

All pixels are at L among the difference correspondence image I ^*a ^*b ^*Variance in the color space on three passages.

(2.2) and at each annular region R _i(p) in, for having those pixels D of certain difference degree with pixel p _i(p), we define its computing formula and are:

Wherein

Remarked pixel

L ^*a ^*b ^*Color feature value, ‖ ‖ ₂Refer to Euclidean distance tolerance.

Represent annular region R _I-1(p) L of all pixels similar in to pixel p ^*a ^*b ^*The color feature vector average, it is defined as follows:

$M_p^{i-1}=\left\{\begin{array}{cc}{\mathrm{\Σ}}_{q^{\′}\∈R_{i-1}\left(p\right)\backslash D_{i-1}\left(p\right)}{I}_{q^{\′}}/{|R_{i-1}\left(p\right)\backslash D_{i-1}\left(p\right)|}_{\mathrm{card}}& \mathrm{or}\\ I_p& \mathrm{if}{D}_{i-1}\left(p\right)=R_{i-1}\left(p\right)\end{array}\right.$

R in the following formula _I-1(p) D _I-1(p) represent annular region R _I-1(p) collection of pixels similar to pixel p in is by set R _I-1(p) with set D _I-1(p) difference obtains, || _CardThe gesture of expression set.For special situation, when at annular region R _I-1(p) in the dissimilar collection of pixels D of pixel p _I-1(p) occupied whole set R _I-1(p) time, that is, and D _I-1(p)=R _I-1(p) time, we make

Promptly

Be exactly the L of pixel p ^*a ^*b ^*The color feature vector value.

(3) based on aforementioned calculation, we provide the computing formula of the present invention about the significance degree S (p) of each pixel p in the image, that is:

$S\left(p\right)=\frac{1}{k}\underset{i=1}{\overset{k}{\mathrm{\Σ}}}\frac{{\left|D_i\left(p\right)\right|}_{\mathrm{card}}}{{\left|R_i\left(p\right)\right|}_{\mathrm{card}}}$

Wherein || _CardThe gesture of expression set.

When using S (p), we need be provided with two parameters, and one is the k value, and promptly we need the annular region among what Fig. 2, and another is the central circular radius r among Fig. 2 ₀, concrete size is determined according to actual conditions.Suggestion r value 3～5, in experiment of the present invention, we make k=3, r ₀=3.

If all adopt S (p) to calculate significance degree to each pixel in the image, so, we just can obtain a width of cloth significance degree distribution plan of this image, Fig. 3 has provided one group of real natural image, and the significance degree distribution plan that obtains by current other three kinds of classic algorithm, simultaneously, we give the significance degree distribution plan that is obtained by computing method of the present invention.As can be seen from Figure 3, the image significance degree distribution plan (Fig. 3 (e)) that is obtained by the present invention not only has identical resolution with original image, and visually has more rationality.

According to an exemplary embodiment of the present invention, be used to realize that computer system of the present invention can comprise, particularly, central processing unit (CPU), storer and I/O (I/O) interface.Computer system usually by I/O interface and display with link to each other such as this type of various input equipments of mouse and keyboard, support circuit can comprise the fast buffer memory of image height, power supply, clock circuit and the such circuit of communication bus.Storer can comprise random access memory (RAM), ROM (read-only memory) (ROM), disc driver, magnetic tape station etc., or their combination.Computer platform also comprises operating system and micro-instruction code.Various process described herein and function can be by the micro-instruction code of operating system execution or the part of application program (or their combination).In addition, various other peripherals can be connected to this computer platform, as additional data storage device and printing device.

Should also be understood that and so the actual connection between the system component (or process steps) may be different, specifically decide on programming mode of the present invention because the assembly and the method step of some construction system described in the accompanying drawing can form of software be realized.Based on the principle of the invention that proposes herein, the ordinary skill of association area it is contemplated that these and similar embodiment or configuration of the present invention.

Claims (1)

1. significance degree computing method of pixel in the image, for each pixel p in the image, calculate as follows:

(1) asks for neighborhood zone in one group of multiple dimensioned scope of pixel p; Wherein,

The neighborhood zone nearest apart from pixel p is a border circular areas R ₀(p), be defined as follows:

R ₀(p)＝{q|0≤||p-q|| ₂≤r ₀，q∈∧}

|| || ₂Expression Euclidean distance tolerance, all pixel in the ∧ presentation video, r ₀Radius for this border circular areas;

Border circular areas R ₀(p) it is the annular region that concentric circles surrounded in the center of circle with the pixel p that the neighborhood zone outside is k, and they are defined as follows:

R _i(p)＝{t|r _i-1≤||p-t|| ₂≤r _i，t∈∧}，i＝1，...，k

r _iIt is ring-like region R _i(p) outer boundary is to the radius of p,

W is the width of image, and H is the height of image, and the interregional radius step-length difference of adjacent annular is constant;

(2) calculate respectively in above-mentioned each neighborhood zone with pixel p and have the pixel of difference degree;

(2.1) at border circular areas R ₀(p) in, has the pixel D of difference degree with pixel p ₀(p) be characterized by:

I wherein _p,

Remarked pixel p,

L ^*a ^*b ^*Color feature value, Wherein

With

Be respectively in the image all pixels at L ^*a ^*b ^*Variance in the color space on three passages;

(2.2) at each annular region R _i(p) in, has the pixel D of difference degree with pixel p _i(p) be characterized by:

I=1 ..., k

Wherein

Remarked pixel

L ^*a ^*b ^*Color feature value,

Or, || _CardThe gesture of expression set, I _{Q '}The L of remarked pixel q ' ^*a ^*b ^*Color feature value;

(3) significance degree of calculating pixel p

Images