CN103679651A - Underwater image enhancement processing method - Google Patents

Abstract

The invention discloses an underwater image enhancement processing method which combines a homomorphic filtering method with a biorthogonal wavelet threshold filtering method. The underwater image enhancement processing method mainly includes the following steps: A), performing logarithm operation on an original image acquired underwater to acquire Z (x,y); B), performing discrete Fourier transformation on Z (x,y) to acquire Z (u,v); C), performing homomorphic filtering processing on Z (u,v) to acquire S (u,v); D), performing Fourier inversion on S (u,v) to acquire s (x,y); E),performing exponent operation on s (x,y) to acquire a processed image g (x,y); F), applying the biorthogonal wavelet threshold filtering method to process the image. By the underwater image enhancement processing method, noise of the image can be removed, contrast ratio of the image is increased, illumination uniformity of a processed underwater image is improved, both balanced filtering and median filtering of a peak signal to noise ratio (PSNR) histogram are improved, underwater image quality is improved, and the underwater image enhancement processing method has certain practical value in improving underwater image quality.

Description

Underwater picture strengthens disposal route

Technical field

The present invention relates to image and process, specifically underwater picture strengthens disposal route.

Background technology

Because Underwater Imaging environment is very complicated, in imaging process, be subject to the impact of several factors, as poor in the optical alignment that floor light light source sends, shooting scene in optical illumination intensity distributions inhomogeneous, picked-up to the bright dark distribution of underwater picture background there is larger difference.In addition, because water has and absorbs and scattering process light, light is subject to strong decay while transmitting in water, the low visibility of underwater camera environment, and the minutia of the underwater picture of acquisition is unintelligible, and contrast is low.Although at present ripe Underwater Imaging technology has been improved underwater picture quality to a certain extent, but still the underwater picture with features such as illumination heterogeneity, poor contrast, signal to noise ratio (S/N ratio) are low cannot meet people's practical application request.Therefore be necessary to adopt corresponding image enhancement processing method, further improve the using value of underwater picture.

Image enhancement processing is a kind of method of improving quality that feature for image and its existing problem are carried out relevant treatment, and main method has: (1) airspace enhancement method, and as mean filter, histogram equalization and medium filtering.(2) frequency domain Enhancement Method, as ideal low-pass filter, wiener filtering and homomorphic filtering.(3) wavelet threshold denoising.At real image, strengthen and process in application, the feature that can exist for image, adopts a kind of or combines several different methods and carry out respective handling.

Histogram equalization is one of typical method in airspace enhancement method, is exactly in fact gradation of image equilibrium, and the method is by pretreatment image after point processing is processed, and the gray-scale value of equilibrium figures picture has identical pixel in each gray level.The grey level histogram of image has embodied the contrast of image and bright dark intuitively, and gray-scale value lacks or is evenly distributed and causes that figure kine bias is dark or partially bright in gray level.The grey value profile of image after histogram equalization is processed is tending towards evenly, and its visual effect is improved, and contrast is improved.The gray-scale value of statistics and analysis image histogram, the gray-scale value statistic histogram function representation of one-dimensional discrete is:

P(S _j)=N _j/N(j=0,1,2,...,L-1)

N in above formula _jbe illustrated in j gray level and comprise number of pixels, N is whole pixels of presentation video.

Medium filtering is the Nonlinear Processing method that suppresses picture noise, and computing is simple, convenient to be realized, and to protect well border be the feature of this image enhancement processing method.There is the moving window of an odd point in medium filtering, in this window, the intermediate value of each pixel grey scale is as the gray scale after being processed by the picture point of filtering in window.

Array [x (i, j)] _{m * N}through window, be A _nmedium filtering after, the output of picture point (i, j) is counted:

$f(i,j)=\underset{A_n(i,j)}{\mathrm{med}}\left[x(i,j)\right]$

A in above formula _n(i, j) represents the field of point (i, j), and it contains n pixel.

Medium filtering effectively in filtering image by the noise scanning or impulse disturbances produces, and can level and smooth other non-pulse noises, reduce distortion, protection image detail.But this filtering algorithm can make image lose the target area of fine rule and fritter.

Figure image intensifying mainly, from containing noisy image, obtains the estimation of original image.Evaluated error is less, shows that figure image intensifying effect is just better.The evaluation criteria of conventional figure image intensifying effect is to weigh with mean square deviation (Mean Squared Error, MSE) and Y-PSNR (Peak Signal to Noise Ratio, PSNR), and its expression formula is formulated as:

$\mathrm{MSE}=\frac{1}{\mathrm{MN}}\left\{\underset{i=0}{\overset{M-1}{\mathrm{\Σ}}}\underset{j=0}{\overset{N-1}{\mathrm{\Σ}}}{[f(i,j)-g(i,j)]}^2\right\}$

$\mathrm{PSNR}=10\lg \left\{\frac{{\left[\max f(i,j)\right]}^2}{\underset{i=0}{\overset{M-1}{\mathrm{\Σ}}}\underset{j=0}{\overset{N-1}{\mathrm{\Σ}}}{[f(i,j)-g(i,j)]}^2}\right\}$

In above formula, M and N are respectively the numbers of x and y directional image pixel, and f (i, j) and g (i, j) are respectively original image and the value of Recovery image on (i, j) point.

These two kinds standards of evaluating noise reduction of mean square deviation and Y-PSNR are objective standard, and they all do not reflect the visual experience of the mankind to picture quality.(MSE) is less for mean square deviation, and (PSNR) is larger for Y-PSNR, and the sharpness of the image after enhancing is higher, and its visual effect is better.

Summary of the invention

The present invention, in order to solve the second-rate problem of underwater picture, provides a kind of underwater picture to strengthen disposal route.

The method is a kind of underwater picture enhancing disposal route of using homomorphic filtering and biorthogonal wavelet threshold filter method to combine.

Homomorphic filtering method is one of frequency domain Enhancement Method.In underwater photographic system because light source collimation is poor, be irradiated to shooting target time uneven illumination even, the underwater picture of picked-up is bright secretly has difference more significantly, and the image section that illumination is stronger shows as brighter, the weak part of illumination shows as darker, and the image detail of this part is fuzzyyer.For the bright dark inhomogeneous underwater picture of this class, can use homomorphic filtering method, the method is in frequency domain, by filter function, to process the low-and high-frequency composition of underwater picture, the brightness range of while pressurized water hypograph, after Nonlinear Processing underwater picture, improve its quality, reduce the inhomogeneous impact of illuminance.

Method of the present invention comprises following key step

A) former underwater picture f (x, y) is carried out to logarithm operation.In imaging model, image f (x, y) can be expressed as the luminance component i (x, y) of light source generation and the product form of reflecting component n (x, y) under water.The relation that is three is: f (x, y)=i (x, y) ' n (x, y), and in ideal conditions, luminance component i (x, y) is a constant, image f (x, y) can distortionless reflection i (x, y).But light field Illumination Distribution is inhomogeneous under water, i (x, y) slowly changes with spatial domain, and on frequency spectrum, its concentration of energy is in low-frequency range.Reflecting component n (x, y) changes with scenery details, and on frequency spectrum, its concentration of energy is at high band.F (x, y) is carried out to logarithm operation:

z(x,y)=lnf(x,y)=ln[i(x,y)′n(x,y)]=lni(x,y)+lnn(x,y)

B) then z (x, y) is carried out to discrete Fourier transformation:

Z(u,v)=F(z(x,y))=F(lni(x,y)+lnn(x,y))=I(u,v)+N(u,v)

C) then use filters H (x, y) to carry out filtering processing to z (x, y), homomorphic filter adopts follow-on Butterworth homomorphic filter, and the mathematic(al) representation of its filter function H (u, v) is:

$H(u,v)=(r_h-r_l)\left[\frac{1}{(1+c*{(D_0/D(u,v))}^{2n})}\right]+r_l$

In formula, r _h>1 representative strengthens the high fdrequency component of image; r _l<1 representative reduces the low frequency component of image; C is constant, gets c=50; D (u, v) is that frequency (u, v) is to filter center (u ₀, v ₀) distance, be expressed as: D (u, v)=[(u-u ₀) ²+ (v-v ₀) ²] ^1/2; D ₀as (u ₀, v ₀)=(0,0), time, the value of D (u, v), represents cutoff frequency, D ₀approximate formula be:

D ₀=a ' median (median (D (u, v))), in formula, a=0.3～0.6

After homomorphic filter is processed, can obtain:

S(u,v)=Z(u,v)×H(u,v)=[I(u,v)+N(u,v)]×H(u,v)=I(u,v)H(u,v)+N(u,v)H(u,v)=F(lni(x,y))H(u,v)+F(lnn(x,y))H(u,v)=I'(u,v)+N'(u,v)

D) S (u, v) is carried out to inverse Fourier transform:

s(x,y)=F ^-1(S(u,v))=F ^-1(I'(u,v)+N'(u,v))=F ^-1(I'(u,v))+F ^-1(N'(u,v))=i'(x,y)+n'(x,y)

E) finally s (x, y) is got to exponent arithmetic, obtain the underwater picture g (x, y) after processing.

F) use biorthogonal wavelet threshold filter method to process underwater picture g (x, y), when retaining the detailed information of underwater picture, the noise of filtering underwater picture.In this experiment, select to there is the biorthogonal wavelet bior1.3 of linear phase feature, and use VisuShrink threshold rule:

(N is the size of picture signal, and s is the standard deviation of image), carries out 3 layers of decomposition to underwater picture g (x, y).

Accompanying drawing explanation

Fig. 1 is the process flow diagram of method of the present invention;

Fig. 2 is the underwater picture before processing;

Fig. 3 is histogram equalization method treatment effect figure;

Fig. 4 is median filtering method treatment effect figure;

Fig. 5 method of the present invention is carried out treatment effect figure to underwater picture.

Embodiment

The Y-PSNR of histogram equalization, medium filtering, the inventive method (PSNR) and mean square deviation (MSE) are as shown in the table.

Table: image enhancement processing method is processed the evaluation index comparison after underwater picture

Image enhancement processing method	PSNR(dB)	MSE
			Histogram equalization	7.5160	1.1521e+004
Medium filtering	31.2067	17.8171
			The inventive method	50.0167	0.7365

By analyzing the feature of underwater picture, the feature such as even for underwater picture uneven illumination, contrast is low, minutia is fuzzy has taked different filtering modes to carry out respective handling.

Comparison diagram 2, Fig. 3, Fig. 4 and Fig. 5, and the data in upper table, can find out, the underwater picture of picked-up has been improved to the homogeneity of the illumination of image with homomorphic filtering, in conjunction with the advantage of biorthogonal wavelet threshold filter filter, remove again the noise of image, improved the contrast of image simultaneously.Underwater picture illumination uniformity after homomorphic filtering is processed improves, its Y-PSNR (PSNR) is all larger than histogram equalization and medium filtering, underwater picture quality is improved, and method of the present invention has certain practical value to improving underwater picture quality.

Claims (5)

1. underwater picture strengthens disposal route, it is characterized in that: use homomorphic filtering and biorthogonal wavelet threshold filter method to combine, comprise following key step

A) by the original image obtaining under water

carry out logarithm operation, obtain z( x, y);

B) then right

carry out discrete Fourier transformation, obtain z( u, v);

C) then right

carry out homomorphic filtering processing, obtain

; Afterwards

D) right

carry out inverse Fourier transform, obtain

;

E) finally right carry out exponent arithmetic,, obtain the image after processing

;

F) use biorthogonal wavelet threshold filter method to image

process.

2. according to the method for claim 1, it is characterized in that: in steps A), right carry out logarithm operation, gained z( x, y) by following formula, expressed:

Wherein, for the luminance component of light source generation,

for reflecting component.

3. according to the method for claim 1, it is characterized in that: homomorphic filtering step C), adopts follow-on Butterworth homomorphic filter, its filter function

mathematic(al) representation be:

>1 representative strengthens the high fdrequency component of image,

<1 representative reduces the low frequency component of image;

for constant, get c=50,

it is frequency

to filter center

distance, be expressed as:

;

to work as

=

time

value, represent cutoff frequency, approximate formula be:

After homomorphic filter is processed, can obtain:

。

4. according to the method for claim 1, it is characterized in that: step D), right

carry out inverse Fourier transform, gained

by following formula, expressed:

。

5. according to the method for claim 1, it is characterized in that: step F), select to there is the biorthogonal wavelet bior1.3 of linear phase feature, and use VisuShrink threshold rule

, to underwater picture

carry out 3 layers of decomposition;

Wherein, N is the size of picture signal,

it is the standard deviation of image.

Images