CN101866481B - Image processing method based on binocular stereoscopic psychological perception - Google Patents

Abstract

The invention belongs to the image processing field and relates to an image processing method based on binocular stereoscopic psychological perception; the method comprises the following steps: read a left eye YUV color image and a right eye YUV color image which are independent relatively, and convert the YUV space of the image into an opponent color space which is in accordance to the nerve centre transmission process by color space conversion; carry out visual field division to the left eye image and the right eye image which carry out color space conversion respectively; exchange the left eye image and the right eye image and reform the left eye image and the right eye image by imitating binocular naso-lateral optic nerve cross functions; process the reformed left eye image and the right eye image respectively by adopting a settable pyramid filter, so as to realize multi-channel frequency resolution of two images; carry out weighting processing to the opponent color spaces of two images respectively according to contrast ratio, sensitivity and weighting coefficients; carry out contrast ratio cover treatment to the opponent color spaces of two images respectively according to the contrast ratio cover parameters. In the invention, image processing is carried out by imitating the process that the human body vision system perceives the stereoscopic image, and the obtained image has real stereoscopic perception.

Description

Image processing method based on the binocular stereoscopic psychological perception

Technical field

The invention belongs to technical field of image processing, relate to a kind of binocular stereo vision model based on human-eye visual characteristic.

Background technology

The mankind can observe the tangible relief three-dimensional scene that has through eyes; The visual characteristic of this and human body eyes has much relations; The about 65mm of mean distance between the eyes; The angle of observing object is slightly different, between the same object that promptly object seen of left eye and right eye are seen fine distinction is arranged, so the mode of describing the scene profile also is not quite similar.Brain according to these two have nuance scene carry out the physiology fusion treatment, produce accurate three-dimensional body perception and the location of this object in scene, have the stereoscopic sensation of the degree of depth exactly.Meeting under the situation of binocular parallax, having only to two eyes provides relatively independent image respectively, just possibly obtain real stereo perception.

Obtain bulk information through vision system human every day, so very important to the research of human-eye visual characteristic and stereoscopic vision.Had a lot of researchists to be engaged in the research work of vision mode in recent years and formed a lot of vision modes, but these models all are to set up to the plane monocular vision, do not consider the characteristic of three-dimensional binocular vision.

Summary of the invention

The present invention on medical science and the basis to human eye working mechanism further investigation, proposes a kind of image processing method that combines binocular stereo vision physiology principle in modern times.For this reason, the present invention adopts following technical scheme:

A kind of image processing method based on the binocular stereoscopic psychological perception comprises the following steps:

The first step, read relatively independent left eye YUV color image and right eye YUV color image, and convert image to meet the nerve center transmission course opposite color space (BW, RG, BY) from yuv space through colour space transformation;

Second step, carry out the visual field to the left-eye image of passing through colour space transformation and eye image respectively and cut apart; Two images all are divided into left and right sides two parts by picture centre; If left part is divided into the a-quadrant; Right part is divided into the B zone, thereby two width of cloth images are divided into left-eye image a-quadrant, left-eye image B zone, eye image a-quadrant and eye image B zone;

The 3rd step, regional and eye image a-quadrant exchange with left-eye image B, reorganization left-eye image and eye image;

The 4th step, employing steerable pyramid wave filter are handled left-eye image and eye image after the reorganization respectively; At first image division is become nondirectional 7 frequency bands radially; Cast out wherein high-frequency sub-band and low frequency sub-band; To keep 5 radially frequency band be divided into 5 grades of subbands of 0-4 from low to high according to frequency, again these 5 grades of subbands are divided the logical subband of band by 0,45,90,135 ° of four direction respectively, realize that the multichannel frequency of two width of cloth images decomposes;

After the 5th step, the process multichannel frequency resolution process; Set subbands at different levels in the opposite color space each opposite color passage W-B of (BW, RG, BY); Contrast sensitivity weight coefficient on R-G and the B-Y carries out weighted respectively to the opposite color space (BW, RG, BY) of two width of cloth images;

The 6th step, set each opposite color passage W-B in opposite color space (BW, RG, BY), the contrast of R-G and B-Y is covered parameter, adopt following formula carry out to the opposite color space (BW, RG, BY) of two width of cloth images respectively degree of comparing cover processing:

$S(f,\mathrm{\θ},n)=\underset{i=0}{\overset{4}{\mathrm{\Σ}}}K\frac{{\left[a(f,\mathrm{\θ},n)\right]}^p}{\underset{\mathrm{\β}}{\mathrm{\Σ}}{\left[a(f,\mathrm{\β},n)\right]}^q+b^2}$

Wherein, wherein, f represents spatial frequency, β, and the direction of the logical subband of band under θ represents, n is that pixel leads to position in the subband corresponding to band, and a function is the pixel value of pixel, and K is a scale parameter, and b is a saturation constant.

As preferred implementation, the step of the first step is following:

At first, change the formula of rgb color into, carry out color transformed according to the YUV color:

R＝Y+1.140V

G＝Y-0.394U-0.581V

B＝Y+2.032U

Change the formula of LMS color again according to rgb color into, carry out color transformed:

L＝-0.0406R+0.7675G+0.1154B

M＝-0.2271R+1.0114G+0.0927B

S＝-0.0427R+0.1562G+0.5536B

Then, change the formula in opposite color space, obtain meeting the opposite color space (BW, RG, BY) of nerve center transmission course through the LMS color:

BW＝0.99L-0.106M-0.094S

RG＝-0.669L+0.742M-0.027S

BY＝-0.212L-0.354M+0.911S。

In the 5th step, choosing of contrast sensitivity weight coefficient sees the following form:

The subband rank	0	1	2	3	4
						W-B	5	19	139	478	496
R-G	154	354	404	184	27
						B-Y	125	332	381	131	28

In the 6th step, choosing of contrast sensitivity weight coefficient sees the following form.

	b	K	P	q
					W-B	6.968	0.29778	2.1158	2
R-G	21.904	0.11379	2.3447	2
					B-Y	13.035	0.07712	2.2788	2

The image processing method that the present invention proposes influences human relief parameter according to some, and the process of imitation human vision system perception stereo-picture is carried out Flame Image Process, thereby the image that obtains has comparatively real stereo perception.On further investigation basis to human vision system, the overall process of eyes perception stereo-picture is imitated and the engineering reproduction, Flame Image Process there is very important meaning, especially the multidimensional image process field in Aero-Space.Be of value to stereo image quality simultaneously and estimate the formulation of respective standard, also be of value to correlative studys such as stereotelevision.

Description of drawings

Fig. 1 binocular stereoscopic psychological perceptual image of the present invention process flow block diagram.

Fig. 2 the present invention adopts Steerable pyramid frequency domain exploded view.

Embodiment

Referring to Fig. 1, binocular stereo vision model performing step of the present invention is following

The first step, colour space transformation

Read relatively independent left eye YUV color image and right eye YUV color image respectively, and convert two width of cloth images to meet the nerve center transmission course opposite color space (BW, RG, BY) from yuv space through colour space transformation.

At first, YUV signal is transformed into brightness value R, G, the B of display, and YUV changes the formula of RGB into] do;

R＝Y+1.140V

G＝Y-0.394U-0.581V

B＝Y+2.032U

Again rgb signal is transformed into L-cones.M-cones in the retina, the response of S-cones cell, the formula that RGB changes LMS into does;

L＝-0.0406R+0.7675G+0.1154B

M＝-0.2271R+1.0114G+0.0927B

S＝-0.0427R+0.1562G+0.5536B

Then,, be transformed into the opposite color space that meets the nerve center transmission course through formula, with piece image with the human psychology approval and component (brightness and the colourity) expression that can quantize.

BW＝0.99L-0.106M-0.094S

RG＝-0.669L+0.742M-0.027S

BY＝-0.212L-0.354M+0.911S

Second step, the visual field are cut apart

Human optic nerve fiber is split into temporo side-looking nerve fibre and nasal side optic nerve fiber; Ganglion neurocyte formation signal in the nethike embrane is divided into two parts according to left and right view field and is sent to optic nerve fiber, gets into temporo side-looking nerve fibre and nasal side optic nerve fiber respectively.

Imitation optic nerve fiber division is cut apart in the visual field of the present invention, carries out the visual field to the left-eye image of passing through colour space transformation and eye image respectively and cuts apart, and two images all are divided into left and right sides Im by picture centre _L(x, y) and Im _R(x, y) two parts are established left half Im _L(x y) is the a-quadrant, right half Im _R(x y) is the B zone, thereby two width of cloth images are divided into left-eye image a-quadrant, left-eye image B zone, four zones in eye image a-quadrant and eye image B zone.The formula that the visual field is cut apart is following:

wherein, X be image pixel vertically the sum

The 3rd step, eyes nasal side chiasma opticum are handled

Eyes nasal side chiasma opticum is handled and is copied optic chiasma function in the human eye.Right eye nasal side optic nerve fiber and left eye nasal side optic nerve fiber carry out coordinated transposition; Right eye nasal side optic nerve fiber and left eye temporo side-looking nerve fibre are formed left tractus opticus and are got into left hemisphere, and left eye nasal side optic nerve fiber and right eye temporo side-looking nerve fibre are formed right tractus opticus and got into right hemisphere.The disposal route in this step of the present invention is: left-eye image B zone and eye image a-quadrant are exchanged reorganization left-eye image and eye image.

The 4th step, multichannel frequency decompose

The multichannel frequency decomposing module becomes picture breakdown the subimage of a plurality of spatial frequencys, direction.Simulate the vision hyperchannel characteristic of human eye with the wave filter of a component layers structure.The effect of bank of filters is a structure of copying the human eye vision perceptual channel; Decompose the pumping signal in the piece image in the different a plurality of spatial frequencys and the subband on the direction, like this each independently subband signal just look like to be image to the excitation of the different perceptual channel of human visual system.

Referring to Fig. 2; The present invention uses Steerable pyramid wave filter to decompose the vision hyperchannel characteristic of match human eye; Utilize the picture breakdown after the steerable pyramid wave filter that is modulated to 0,45,90,135 degree is respectively recombinated each to be 7 radially frequency bands, as shown in Figure 2.At first radially cast out high-frequency sub-band and low frequency sub-band in the frequency band at 7 that divide; Keep remaining 5 radially frequency bands; Then with these 5 frequency bands radially; To keep 5 radially frequency band be divided into 5 other subbands of level of 0-4 from low to high according to frequency, divide the logical subbands of band by 0,45 °, 90 °, 135 ° four directions respectively, realize that the multichannel frequency of two width of cloth images decomposes.

The 5th step, contrast sensitivity weighting

After the multichannel frequency resolution process, change into contrast to the output of each subband, further utilize known visual contrast sensitivity and contrast to cover knowledge and approach human-eye visual characteristic.The corresponding numerical value of subband, the contrast sensitivity function of all subbands combine and meet the CSF curved surface that the psychology of vision experiment obtains, with five of five coefficients correspondences radially the weighting of frequency band approach the contrast sensitivity function.Present embodiment subbands at different levels are each opposite color passage W-B of (BW, RG, BY) in the opposite color space, and the selection of the contrast sensitivity weight coefficient on R-G and the B-Y is as shown in the table.

The contrast sensitivity weight table

The subband rank	0	1	2	3	4
						W-B	5	19	139	478	496
R-G	154	354	404	184	27
						B-Y	125	332	381	131	28

The 6th step, contrast are covered

Can think that it is that the response of excitor nerve unit receives other neuron to unite the non-linear inhibition of response that contrast is covered.Simulating contrast more intuitively, to cover mechanism method all be to get maximal value usually, make signal amplitude only after surpassing certain threshold value the influence of its performance just can embody.Use certain normalized model will cover certainly and cover the way that visual characteristic joins together to consider with adjacency channel.Can simulate the shielding effect between the equidirectional excitation like this, can also embody covering each other between the different excitation of direction.The present invention adopts following formula degree of comparing to cover:

$S(f,\mathrm{\θ},n)=\underset{i=0}{\overset{4}{\mathrm{\Σ}}}K\frac{{\left[a(f,\mathrm{\θ},n)\right]}^p}{\underset{\mathrm{\β}}{\mathrm{\Σ}}{\left[a(f,\mathrm{\β},n)\right]}^q+b^2}$

Wherein, f represents spatial frequency, β, and the direction of the logical subband of band under θ represents, n is that pixel leads to position in the subband corresponding to band, and a function is the pixel value of pixel, and K is a scale parameter, and b is saturation constant, coefficient selecting such as following table.In the denominator to same radial different directions sub-band coefficients square summation simulated the interaction between equidirectional and the different directions signal.

Contrast is covered parameter list

	b	K	P	q
					W-B	6.968	0.29778	2.1158	2
R-G	21.904	0.11379	2.3447	2
					B-Y	13.035	0.07712	2.2788	2

Claims (3)

1. the image processing method based on the binocular stereoscopic psychological perception comprises the following steps:

The first step, read relatively independent left eye YUV color image and right eye YUV color image, and convert image to meet the nerve center transmission course opposite color space (BW, RG, BY) from yuv space through colour space transformation;

Second step, carry out the visual field to the left-eye image of passing through colour space transformation and eye image respectively and cut apart; Two images all are divided into left and right sides two parts by picture centre; If left part is divided into the a-quadrant; Right part is divided into the B zone, thereby two width of cloth images are divided into left-eye image a-quadrant, left-eye image B zone, eye image a-quadrant and eye image B zone;

The 3rd step, regional and eye image a-quadrant exchange with left-eye image B, reorganization left-eye image and eye image;

The 4th step, employing steerable pyramid wave filter are handled left-eye image and eye image after the reorganization respectively; At first image division is become nondirectional 7 frequency bands radially; Cast out wherein high-frequency sub-band and low frequency sub-band; To keep 5 radially frequency band be divided into 5 grades of subbands of 0-4 from low to high according to frequency, again these 5 grades of subbands are divided the logical subbands of band by 0 °, 45 °, 90 °, 135 ° four directions respectively, realize that the multichannel frequency of two width of cloth images decomposes;

After the 5th step, the process multichannel frequency resolution process; Set subbands at different levels in the opposite color space each opposite color passage W-B of (BW, RG, BY); Contrast sensitivity weight coefficient on R-G and the B-Y carries out weighted respectively to the opposite color space (BW, RG, BY) of two width of cloth images;

Each opposite color passage W-B in the 6th step, setting opposite color space (BW, RG, BY), the contrast of R-G and B-Y is covered parameter, and contrast is covered choosing of parameter and is seen the following form:

? b K P q W-B 6.968 0.29778 2.1158 2 R-G 21.904 0.11379 2.3447 2 B-Y 13.035 0.07712 2.2788 2

Adopt following formula carry out to the opposite color space (BW, RG, BY) of two width of cloth images respectively degree of comparing cover processing:

Wherein, wherein, f represents spatial frequency, β, and the direction of the logical subband of band under θ represents, n is that pixel leads to position in the subband corresponding to band, and a function is the pixel value of pixel, and K is a scale parameter, and b is a saturation constant.

2. the image processing method based on the binocular stereoscopic psychological perception according to claim 1 is characterized in that, the step of the first step is following:

At first, change the formula of rgb color into, carry out color transformed according to the YUV color:

R＝Y+1.140V

G＝Y-0.394U-0.581V

B＝Y+2.032U

Change the formula of LMS color again according to rgb color into, carry out color transformed:

L＝-0.0406R+0.7675G+0.1154B

M＝-0.2271R+1.0114G+0.0927B

S＝-0.0427R+0.1562G+0.5536B

Then, change the formula in opposite color space, obtain meeting the opposite color space (BW, RG, BY) of nerve center transmission course through the LMS color:

BW＝0.99L-0.106M-0.094S

RG＝-0.669L+0.742M-0.027S

BY＝-0.212L-0.354M+0.911S。

3. the image processing method based on the binocular stereoscopic psychological perception according to claim 1 is characterized in that, in the 5th step, choosing of contrast sensitivity weight coefficient sees the following form:

The subband rank 0 1 2 3 4 W-B 5 19 139 478 496 R-G 154 354 404 184 27 B-Y 125 332 381 131 28

。

Images