An Improved Biomimetic Image Processing Method - Semantic Scholar
An Improved Biomimetic Image Processing Method Chen Chen, Weijun Li*, and Liang Chen Lab of Artificial Neural Networks, Institute of Semiconductors, Chinese Academy of Sciences, Beijing, China Tel.: +86 010 8230 4336 {eunicechen,wjli,achenliang}@semi.ac.cn
Abstract. Biomimetics is a rapidly developing discipline and has been suggested applicable in machine vision and image processing because human vision system has almost evolved to be perfect. Previously proposed BA+DRF method is a biomimetic image processing method which improves images quality effectively on the basis of the brightness adaption and disinhibitory properties of concentric receptive field (DRF). However, BA+DRF is not automatic and dynamic leading to the lack of practicability. This paper proposes an improved biomimetic image processing method, the parameterized LDRF method, to make BA+DRF method more adaptive and dynamic. Parameterized LDRF method constructed a parameterized logarithmic model to automatically enhance the image’s global quality and constructed a model to dynamically adjust the gain factor which is used in improving the image’s local quality. The experimental results have proved its ability of enhancing the image quality with keeping details. The improved biomimetic image processing method is applicable and automatic. Keywords: Biomimetic, visual characteristics, logarithm, disinhibitory properties of concentric receptive field (DRF), parameterized dynamic, gain factor.
1 Introduction Biomimetics is a rapidly developing discipline. More recently, biomimetics have been suggested applicable in machine vision systems, image processing, and data converters. Human visual system is the most effective image processor and has almost evolved to be perfect. Therefore, visual characteristics and its application in image processing have aroused much interest among researchers. Since 1970 when Robison did his research on monkeys’ visual neural system, scientists began to deeply study the physiology of visual mechanisms and explored new image processing methods on the basis of bionic vision [1]. For example, Land E.H. proposed image enhancement algorithm based on retinex theory [2-4] and Zi Fang applied vision bionics in the design of multi-waveband imaging guidance head [5]. Researches show that applying physiological visual mechanisms in image processing technology has efficiently improved speed and quality of image processing algorithms. *
Corresponding author.
H. Deng et al. (Eds.): AICI 2011, Part II, LNAI 7003, pp. 246–254, 2011. © Springer-Verlag Berlin Heidelberg 2011
An Improved Biomimetic Image Processing Method
247
In 2010, Xiaoxian Jin introduced BA+DRF method to imitate the brightness adaption feature of vision characteristics and the disinhibitory properties of retinal neuron receptive field [6, 11]. BA+DRF method can be used for image processing and produce better results. However, BA+DRF method has many numerical variables that require researchers to assign a value to each parameter for every image. Therefore, it is not able to improve images quality automatically and dynamically which results in lacking practicability. In order to alleviate these problems, this paper proposes an improved biomimetic image processing method based on BA+DRF method. We constructed models for some parameters that are used in BA+DRF and named this method as Parameterized LDRF method that can apparently improve the practicability. In section 2, we review the visual characteristics and BA+DRF method. In section 3, we introduce the improved global image enhancement method that embeds a parameter model and present the experimental results. In section 4, a biomimetic local image enhancement method that includes a parameter model is demonstrated and the experimental results are shown. In section 5, we get to the conclusion of this parameterized LDRF method.
2 Visual Characteristic and BA+DRF Method 2.1 Visual Characteristic Because digital images are displayed as a discrete set of intensities, the eye’s ability to discriminate between different intensity levels is an important consideration in image processing technology. Experimental evidence indicates that subjective brightness is a logarithmic function of the light intensity incident on the eye. Fig. 1, a plot of light intensity versus subjective brightness, illustrates this characteristic. This non-linear brightness adaptation happened at the beginning of visual system. [7, 8] In 1991, Chaoyi Li, [12, 13] discovered that there was a large range of disinhibitory region outside the classical receptive field, which is very insensitive, but acts as an important role in transmitting the large-area brightness and the brightness gradients (Fig. 2). Because the activity of this region can offset the antagonism between the peripheral region and the central region, it will also compensate the decline of the low frequency image components in the classical receptive field structure. Li gave a new model of Tri-Guassians (as in (1)) to imitate the disinhibitory properties of concentric receptive field (DRF). 2
G ( x, y ) = A1 exp(−
x +y 2σ 1
2
2
2
) − A2 exp( −
x +y 2σ 2
2
2
2
) + A3 exp(−
x +y 2σ 3
2
2
)
(1)
Where A1 , A2 , A3 denote the response intensity of the central, peripheral, and border region, σ 1 , σ 2 , σ 3 denote the corresponding range.
248
C. Chen, W. Li, and L. Chen
Fig. 1. Logarithm of intensity
Fig. 2. Disinhibitory Properties of Concentric Receptive Field (DRF)
2.2 BA+DRF Method
BA+DRF method [6, 11] implements brightness adaptation function and DRF function in image enhancement. First, a global logarithmic transform (as in (2)) is carried out on the image. Second, the DRF Tri-Gaussians model combined with the bi-lateral gaussian filter (as in (3), (4), (5)) is used to enhance the local contrast of the image. When intensity of pixel is lighter than its neighbor’s intensities which are perceived by human vision, the pixel’s intensity value will be increased and when intensity of pixel is darker than its neighbor, the pixel’s intensity value will be decreased. Finally, linear transformation is utilized to restore image. I g ( x , y ) = log( I ( x, y ) + 1) / log(256)
(2)
I lin ( x, y ) = parak • ( I g ( x, y ) − IV ( x, y)) + IV ( x, y )
(3)
An Improved Biomimetic Image Processing Method
IV ( x, y) =
M
249
M
i , j =− M
GRGV I g ( xi , y j )
i , j =− M
GRGV
(4)
2
GV (Ig (x, y), Ig (xi , y j )) = exp(−
(Ig (x, y) − Ig (xi , y j )) 2σV
2
)
(5)
Where parak is a gain factor and its value is positive constant. I ( x , y ) is the pixel value of the original image. I g ( x , y ) is the pixel value of globally enhanced images.
I v ( x, y ) is the subjective intensity of pixel ( x, y ) . GR represents the DRF Tri-Guassians model shown in (1) Gv is the guassian function that computes similarity between center pixel value and its surrounding. σ v is the scale parameter to adjust decay rate of guassian function.
3 Parameterized Logarithmic Model in Global Enhancement and Experimental Results Parameterized logarithmic model is the first step of parameterized LDRF method to globally enhance images contrast. Considering logarithm of intensity which has the fundamental of human vision system and each image condition, we compute images intensity by parameterized logarithmic model (as in (6)-(9)) I ( x, y ) = c( k ) • [log( I ( x, y ) + k + 1) / log( m( k )) − t ( k )] g
(6)
m ( k ) = 256 + k
(7)
t ( k ) = log( k + 1) / log( m( k ))
(8)
c ( k ) = 1 / (1 − t ( k ))
(9)
Where I ( x , y ) is the pixel value of the original image and I g ( x , y ) is the pixel value of the globally enhanced image. m ( k ), t ( k ), c ( k ) are parameters that are used for adjusting pixels intensity range automatically according to the condition of images themselves. k is a parameter that depends on original image intensity condition. Here, we use threshold value (thresh) to determine k which is shown by the following law:
250
C. Chen, W. Li, and L. Chen
If {sum(histogram(1:k)) / sum(histogram(:)) thresh}, the value of k is used in (6)-(9). The parameterized logarithmic model can enhance the image automatically and adaptively to the image’s own condition which is very useful in avoiding enhancing images too light. Based on the properties of the logarithmic function, the global nonlinear transform of the image can efficiently boost the intensity of the dark area of the image, but will compress the dynamic range of the intensity that results in the lack of the image’s details. The experimental results of global image enhancement are shown in Fig. 3 (b) to Fig. 6 (b). For example, comparing Fig. 3 (b) with Fig. 3 (a), the light contrast in the building’s door part of images becomes larger, while the light contrast in the sky part of images becomes smaller. The same changing also happened in Fig. 4 (b) to Fig. 6 (b).
4 Parameterized DRF Model in Local Enhancement and Experimental Results Parameterized DRF model is the second step of parameterized LDRF method which is to locally improve the quality of the image. After we studied statistical differencing [9] and Wallis statistical difference operator [10], we construct a model to compute gain factor parak which is just a constant in BA+DRF method. In the parameterized LDRF method, the model of paraK is presented in (10), (11), (12). parak ( x, y ) =
Amax * Dd Amax * D ( x, y ) + Dd
Dmax = max{D ( x, y )} Dd =
Amin * Amax * Dmax Amax − Amin
(10) (11) (12)
Where parak ( x, y ) is a gain factor function and its value is positive. D ( x, y ) is a standard deviation that is calculated in the w*w field around the ( x, y ) pixel. , Amax is the smallest and largest values of gain factor, according to experience, the value are 1 and 3 and these values are applicable for most of images. Equation (2) to (5) and (10) to (12) compose the parameterized DRF method. According to (10) (11) (12), we can improve the image quality more dependent on the local contrast information. If the local intensity contrast is larger, the gain factor will be smaller and if the local intensity contrast is smaller, the gain factor will be larger. In addition, (12) can limited gain factor in the range of [ Amin , Amax ] . Therefore, Amin
parameterized DRF model is able to improve image quality locally and dynamically.
An Improved Biomimetic Image Processing Method
251
Fig. 3. Building images, (a): Original image, (b): global enhanced image, (c): local enhanced image, (d): parak(x, y)
Fig. 4. Office images, (a): Original image, (b): global enhanced image, (c): local enhanced image, (d): parak(x, y)
252
C. Chen, W. Li, and L. Chen
Fig. 5. Tree images, (a): Original image, (b): global enhanced image, (c): local enhanced image, (d): parak(x, y)
Fig. 6. Facial images, (a): Original image, (b): global enhanced image, (c): local enhanced image, (d): parak(x, y)
An Improved Biomimetic Image Processing Method
253
The combination of Parameterized Logarithmic model and Parameterized DRF model better imitates the varying response of human eye at large light deviation versus small light deviation. These two models compose the parameterized LDRF method. We used parameterized LDRF method to enhance four images. The experimental results are shown in Fig. 3 to Fig. 6. In Fig. 3 to Fig. 6, the image’s quality is dynamically improved in the range of [1, 3]. When gain factor is equal to 1, it means that the local contrast of the image has no need to be boosted while the value is larger, the contrast of local field will be greater. When gain factor is larger than 3, the enhanced image will be too light to lose some intensity information. Fig. 3 (c), Fig. 4 (c), Fig. 5 (c), and Fig. 6 (c) give the value of gain factor parak ( x, y ) of different images which show that various images is assigned with various gain factors and they correspond to the local region of images. Experimental results prove that parameterized LDRF method is able to enhance some area smoothly and other area sharply that is to mean this kind of biomimetic image enhancement method can improve image quality adaptively, automatically and dynamically.
5 Conclusion This paper proposed an improved biomimetic image processing method based on the BA+DRF method to imitate brightness adaption and Disinhibitory Properties of Concentric Receptive. This improved biomimetic method, parameterized LDRF method, has constructed parameter models to make biomimetic method more adaptive, automatic and dynamic. The experimental results have proved its applicability and practicability. Visual bionic is gradually developing from theoretical study to practical application. The research on biomimetic image processing method will be helpful both in bettering image processing technologies and in boosting the deeply understanding of human visual system. In the future, we will continue contribute our effort to study biomimetic image processing method and push it developing. Acknowledgments. National Natural Science Foundation of China (No. 90920013) supported this work.
References 1. Wang, H.-m., Li, Y.-j., Zhang, K.: Application of bio-vision bionics on computer vision. Application Research of Computers 26(3), 1157–1159 (2009) 2. Kolb, H.: How the Retina Works. American Scientist 91(1) (2003) 3. Land, E.H.: The Retina Theory of Color Vision. Scientific American 237(6), 108–129 (1997) 4. Jobson, D.J., Rahman, Z., Woodell, G.A.: Properties and Performance of a Center / Surround Retinex. IEEE Transaction on Image Processing 6(3), 451–462 (1997) 5. Zi, F., Li, Y.-j., Zhao, D.-w., Zhang, K.: Vision Bionics and Its Application on Design of Multi-Wavehand Imaging Guidance Head. Control Technology 27(1), 55–56 (2008) 6. Jin, X., Li, W., Wang, S.: An algorithm for Biomimetic Image Enhancement Based on Human Visual Property. CAD&CG 22(3), 534–537 (2010)
254
C. Chen, W. Li, and L. Chen
7. Gonzalez, R.C., Woods, R.E.: Digital Image Processing, 2nd edn. Prentice Hall, Englewood Cliffs (2002); ISBN 978-7-121-04398-7 8. Pratt, W.K.: Digital Image Processing, 4th edn. Wiley, Chichester (2009); ISBN 978-7111-28968-5 9. Rosenfeld, A.: Picture Processing by Computer. Academic Press, New York (1969) 10. Wallis, R.H.: An Approach for the Space Variant Restoration and Enhancement of Images. In: Proc. Symposium on Current Mathematical Problems in Image Science, Monterey, CA (November 1976) 11. Jin, X.: Study on Biomimetic Processing Method of Face Image. Institute of Semiconductors, Chinese Academey of Sciences, Beijing (2010) 12. Qiu, F., Li, C.: Mathematical Stimulation of Disinhibitory Properties of Concentric Receptive Field. Acta Biophysica Sinica 11(2), 214–220 (1995) 13. Li, C.Y., Pei, X., et al.: Role of the extensive area outside the x-cell receptive field in brightness information transmission. Vision Research 31(9), 1529–1540 (1991)
Abstract. Biomimetics is a rapidly developing discipline and has been suggested applicable in machine vision and image processing because human vision system has almost evolved to be perfect. Previously proposed BA+DRF method is a biomimetic image processing method which improves images quality effectively on the basis of the brightness adaption and disinhibitory properties of concentric receptive field (DRF). However, BA+DRF is not automatic and dynamic leading to the lack of practicability. This paper proposes an improved biomimetic image processing method, the parameterized LDRF method, to make BA+DRF method more adaptive and dynamic. Parameterized LDRF method constructed a parameterized logarithmic model to automatically enhance the image’s global quality and constructed a model to dynamically adjust the gain factor which is used in improving the image’s local quality. The experimental results have proved its ability of enhancing the image quality with keeping details. The improved biomimetic image processing method is applicable and automatic. Keywords: Biomimetic, visual characteristics, logarithm, disinhibitory properties of concentric receptive field (DRF), parameterized dynamic, gain factor.
1 Introduction Biomimetics is a rapidly developing discipline. More recently, biomimetics have been suggested applicable in machine vision systems, image processing, and data converters. Human visual system is the most effective image processor and has almost evolved to be perfect. Therefore, visual characteristics and its application in image processing have aroused much interest among researchers. Since 1970 when Robison did his research on monkeys’ visual neural system, scientists began to deeply study the physiology of visual mechanisms and explored new image processing methods on the basis of bionic vision [1]. For example, Land E.H. proposed image enhancement algorithm based on retinex theory [2-4] and Zi Fang applied vision bionics in the design of multi-waveband imaging guidance head [5]. Researches show that applying physiological visual mechanisms in image processing technology has efficiently improved speed and quality of image processing algorithms. *
Corresponding author.
H. Deng et al. (Eds.): AICI 2011, Part II, LNAI 7003, pp. 246–254, 2011. © Springer-Verlag Berlin Heidelberg 2011
An Improved Biomimetic Image Processing Method
247
In 2010, Xiaoxian Jin introduced BA+DRF method to imitate the brightness adaption feature of vision characteristics and the disinhibitory properties of retinal neuron receptive field [6, 11]. BA+DRF method can be used for image processing and produce better results. However, BA+DRF method has many numerical variables that require researchers to assign a value to each parameter for every image. Therefore, it is not able to improve images quality automatically and dynamically which results in lacking practicability. In order to alleviate these problems, this paper proposes an improved biomimetic image processing method based on BA+DRF method. We constructed models for some parameters that are used in BA+DRF and named this method as Parameterized LDRF method that can apparently improve the practicability. In section 2, we review the visual characteristics and BA+DRF method. In section 3, we introduce the improved global image enhancement method that embeds a parameter model and present the experimental results. In section 4, a biomimetic local image enhancement method that includes a parameter model is demonstrated and the experimental results are shown. In section 5, we get to the conclusion of this parameterized LDRF method.
2 Visual Characteristic and BA+DRF Method 2.1 Visual Characteristic Because digital images are displayed as a discrete set of intensities, the eye’s ability to discriminate between different intensity levels is an important consideration in image processing technology. Experimental evidence indicates that subjective brightness is a logarithmic function of the light intensity incident on the eye. Fig. 1, a plot of light intensity versus subjective brightness, illustrates this characteristic. This non-linear brightness adaptation happened at the beginning of visual system. [7, 8] In 1991, Chaoyi Li, [12, 13] discovered that there was a large range of disinhibitory region outside the classical receptive field, which is very insensitive, but acts as an important role in transmitting the large-area brightness and the brightness gradients (Fig. 2). Because the activity of this region can offset the antagonism between the peripheral region and the central region, it will also compensate the decline of the low frequency image components in the classical receptive field structure. Li gave a new model of Tri-Guassians (as in (1)) to imitate the disinhibitory properties of concentric receptive field (DRF). 2
G ( x, y ) = A1 exp(−
x +y 2σ 1
2
2
2
) − A2 exp( −
x +y 2σ 2
2
2
2
) + A3 exp(−
x +y 2σ 3
2
2
)
(1)
Where A1 , A2 , A3 denote the response intensity of the central, peripheral, and border region, σ 1 , σ 2 , σ 3 denote the corresponding range.
248
C. Chen, W. Li, and L. Chen
Fig. 1. Logarithm of intensity
Fig. 2. Disinhibitory Properties of Concentric Receptive Field (DRF)
2.2 BA+DRF Method
BA+DRF method [6, 11] implements brightness adaptation function and DRF function in image enhancement. First, a global logarithmic transform (as in (2)) is carried out on the image. Second, the DRF Tri-Gaussians model combined with the bi-lateral gaussian filter (as in (3), (4), (5)) is used to enhance the local contrast of the image. When intensity of pixel is lighter than its neighbor’s intensities which are perceived by human vision, the pixel’s intensity value will be increased and when intensity of pixel is darker than its neighbor, the pixel’s intensity value will be decreased. Finally, linear transformation is utilized to restore image. I g ( x , y ) = log( I ( x, y ) + 1) / log(256)
(2)
I lin ( x, y ) = parak • ( I g ( x, y ) − IV ( x, y)) + IV ( x, y )
(3)
An Improved Biomimetic Image Processing Method
IV ( x, y) =
M
249
M
i , j =− M
GRGV I g ( xi , y j )
i , j =− M
GRGV
(4)
2
GV (Ig (x, y), Ig (xi , y j )) = exp(−
(Ig (x, y) − Ig (xi , y j )) 2σV
2
)
(5)
Where parak is a gain factor and its value is positive constant. I ( x , y ) is the pixel value of the original image. I g ( x , y ) is the pixel value of globally enhanced images.
I v ( x, y ) is the subjective intensity of pixel ( x, y ) . GR represents the DRF Tri-Guassians model shown in (1) Gv is the guassian function that computes similarity between center pixel value and its surrounding. σ v is the scale parameter to adjust decay rate of guassian function.
3 Parameterized Logarithmic Model in Global Enhancement and Experimental Results Parameterized logarithmic model is the first step of parameterized LDRF method to globally enhance images contrast. Considering logarithm of intensity which has the fundamental of human vision system and each image condition, we compute images intensity by parameterized logarithmic model (as in (6)-(9)) I ( x, y ) = c( k ) • [log( I ( x, y ) + k + 1) / log( m( k )) − t ( k )] g
(6)
m ( k ) = 256 + k
(7)
t ( k ) = log( k + 1) / log( m( k ))
(8)
c ( k ) = 1 / (1 − t ( k ))
(9)
Where I ( x , y ) is the pixel value of the original image and I g ( x , y ) is the pixel value of the globally enhanced image. m ( k ), t ( k ), c ( k ) are parameters that are used for adjusting pixels intensity range automatically according to the condition of images themselves. k is a parameter that depends on original image intensity condition. Here, we use threshold value (thresh) to determine k which is shown by the following law:
250
C. Chen, W. Li, and L. Chen
If {sum(histogram(1:k)) / sum(histogram(:)) thresh}, the value of k is used in (6)-(9). The parameterized logarithmic model can enhance the image automatically and adaptively to the image’s own condition which is very useful in avoiding enhancing images too light. Based on the properties of the logarithmic function, the global nonlinear transform of the image can efficiently boost the intensity of the dark area of the image, but will compress the dynamic range of the intensity that results in the lack of the image’s details. The experimental results of global image enhancement are shown in Fig. 3 (b) to Fig. 6 (b). For example, comparing Fig. 3 (b) with Fig. 3 (a), the light contrast in the building’s door part of images becomes larger, while the light contrast in the sky part of images becomes smaller. The same changing also happened in Fig. 4 (b) to Fig. 6 (b).
4 Parameterized DRF Model in Local Enhancement and Experimental Results Parameterized DRF model is the second step of parameterized LDRF method which is to locally improve the quality of the image. After we studied statistical differencing [9] and Wallis statistical difference operator [10], we construct a model to compute gain factor parak which is just a constant in BA+DRF method. In the parameterized LDRF method, the model of paraK is presented in (10), (11), (12). parak ( x, y ) =
Amax * Dd Amax * D ( x, y ) + Dd
Dmax = max{D ( x, y )} Dd =
Amin * Amax * Dmax Amax − Amin
(10) (11) (12)
Where parak ( x, y ) is a gain factor function and its value is positive. D ( x, y ) is a standard deviation that is calculated in the w*w field around the ( x, y ) pixel. , Amax is the smallest and largest values of gain factor, according to experience, the value are 1 and 3 and these values are applicable for most of images. Equation (2) to (5) and (10) to (12) compose the parameterized DRF method. According to (10) (11) (12), we can improve the image quality more dependent on the local contrast information. If the local intensity contrast is larger, the gain factor will be smaller and if the local intensity contrast is smaller, the gain factor will be larger. In addition, (12) can limited gain factor in the range of [ Amin , Amax ] . Therefore, Amin
parameterized DRF model is able to improve image quality locally and dynamically.
An Improved Biomimetic Image Processing Method
251
Fig. 3. Building images, (a): Original image, (b): global enhanced image, (c): local enhanced image, (d): parak(x, y)
Fig. 4. Office images, (a): Original image, (b): global enhanced image, (c): local enhanced image, (d): parak(x, y)
252
C. Chen, W. Li, and L. Chen
Fig. 5. Tree images, (a): Original image, (b): global enhanced image, (c): local enhanced image, (d): parak(x, y)
Fig. 6. Facial images, (a): Original image, (b): global enhanced image, (c): local enhanced image, (d): parak(x, y)
An Improved Biomimetic Image Processing Method
253
The combination of Parameterized Logarithmic model and Parameterized DRF model better imitates the varying response of human eye at large light deviation versus small light deviation. These two models compose the parameterized LDRF method. We used parameterized LDRF method to enhance four images. The experimental results are shown in Fig. 3 to Fig. 6. In Fig. 3 to Fig. 6, the image’s quality is dynamically improved in the range of [1, 3]. When gain factor is equal to 1, it means that the local contrast of the image has no need to be boosted while the value is larger, the contrast of local field will be greater. When gain factor is larger than 3, the enhanced image will be too light to lose some intensity information. Fig. 3 (c), Fig. 4 (c), Fig. 5 (c), and Fig. 6 (c) give the value of gain factor parak ( x, y ) of different images which show that various images is assigned with various gain factors and they correspond to the local region of images. Experimental results prove that parameterized LDRF method is able to enhance some area smoothly and other area sharply that is to mean this kind of biomimetic image enhancement method can improve image quality adaptively, automatically and dynamically.
5 Conclusion This paper proposed an improved biomimetic image processing method based on the BA+DRF method to imitate brightness adaption and Disinhibitory Properties of Concentric Receptive. This improved biomimetic method, parameterized LDRF method, has constructed parameter models to make biomimetic method more adaptive, automatic and dynamic. The experimental results have proved its applicability and practicability. Visual bionic is gradually developing from theoretical study to practical application. The research on biomimetic image processing method will be helpful both in bettering image processing technologies and in boosting the deeply understanding of human visual system. In the future, we will continue contribute our effort to study biomimetic image processing method and push it developing. Acknowledgments. National Natural Science Foundation of China (No. 90920013) supported this work.
References 1. Wang, H.-m., Li, Y.-j., Zhang, K.: Application of bio-vision bionics on computer vision. Application Research of Computers 26(3), 1157–1159 (2009) 2. Kolb, H.: How the Retina Works. American Scientist 91(1) (2003) 3. Land, E.H.: The Retina Theory of Color Vision. Scientific American 237(6), 108–129 (1997) 4. Jobson, D.J., Rahman, Z., Woodell, G.A.: Properties and Performance of a Center / Surround Retinex. IEEE Transaction on Image Processing 6(3), 451–462 (1997) 5. Zi, F., Li, Y.-j., Zhao, D.-w., Zhang, K.: Vision Bionics and Its Application on Design of Multi-Wavehand Imaging Guidance Head. Control Technology 27(1), 55–56 (2008) 6. Jin, X., Li, W., Wang, S.: An algorithm for Biomimetic Image Enhancement Based on Human Visual Property. CAD&CG 22(3), 534–537 (2010)
254
C. Chen, W. Li, and L. Chen
7. Gonzalez, R.C., Woods, R.E.: Digital Image Processing, 2nd edn. Prentice Hall, Englewood Cliffs (2002); ISBN 978-7-121-04398-7 8. Pratt, W.K.: Digital Image Processing, 4th edn. Wiley, Chichester (2009); ISBN 978-7111-28968-5 9. Rosenfeld, A.: Picture Processing by Computer. Academic Press, New York (1969) 10. Wallis, R.H.: An Approach for the Space Variant Restoration and Enhancement of Images. In: Proc. Symposium on Current Mathematical Problems in Image Science, Monterey, CA (November 1976) 11. Jin, X.: Study on Biomimetic Processing Method of Face Image. Institute of Semiconductors, Chinese Academey of Sciences, Beijing (2010) 12. Qiu, F., Li, C.: Mathematical Stimulation of Disinhibitory Properties of Concentric Receptive Field. Acta Biophysica Sinica 11(2), 214–220 (1995) 13. Li, C.Y., Pei, X., et al.: Role of the extensive area outside the x-cell receptive field in brightness information transmission. Vision Research 31(9), 1529–1540 (1991)
