Recently, Xu et al. Image processing shows an important part of collecting information on brain images. In their work, the input data and output data are low-quality images and high-quality images, respectively. The final optimized result is close to the original image \({I_0}\) and satisfies the GW assumption, that is, the final result is close to 1/2. Histogram equalization(HE) is a common method in image enhancement, which does not require parameter settings of external factors and can effectively enhance image contrast [22]. (Pdf) Importance of Image Enhancement Techniques in Color Image According to [9], SSR model can be described as: where \({I_i}\) is the image distributed in the i-th color band, \({R_i}(x,y)\) is the enhancement result, * means convolution operation, F(x,y) represents the convolution kernel function and the formula is as follows: where k needs to satisfy \(\iint \limits F(x,y) \mathrm {d}x \mathrm {d}y=1\). In particular, FLM causes a bad effect on image details preservation. 6, we can see that the center/surround-based Retinex algorithm can effectively enhance the contrast of the image. Typical methods are DHE [39] and BPDHE [40]. IEEE Trans Image Process 9(3):505510, Lin S, Wong C, Jiang G, Rahman M, Ren T, Kwok N, Shi H, Yu Y-H, Wu T (2016) Intensity and edge based adaptive unsharp masking filter for color image enhancement. Although the local details of the image enhancement method based on local histogram equalization will become richer, the average brightness will usually be greatly affected. IEEE Trans on Image Process 29:50225037, Tang M, Xie F, Zhang R, Jiang Z, Bovik AC (2020) A local flatness based variational approach to retinex. EnlightenGAN uses attention-guided U-Net as a generator, and uses dual discriminators (global and local) to balance global and local low-light enhancement. Finally, attention-based multi-scale feature aggregation is proposed for image enhancement. In addition, the contrast term is described as follows: where \({\omega (x,y)}\) is the distance function, and its value decreases monotonously with the increase of the distance between pixels. [52] proved that the threshold mechanism has very little contribution to the model algorithm, so (5) can be simplified as: For the path-based Retinex algorithm, the researchers have focused on the selection of the pixel path, and lots of related work has gradually emerged. In implementation, the enhancement image can be obtained by the negative of the time matrix. In this paper, considering the limitation of space, we only summarize a few main parts of the method of histogram modification. Image enhancement is considered as one of the most important techniques in image research. The frequency domain image enhancement is to convert the image to the frequency domain in a certain mathematical function such as Fourier transform (FT), discrete cosine transform (DCT) and discrete wavelet transform (DWT), then perform image processing based on the unique properties of the frequency domain, and finally convert it to the original image space. Siam J Imaging Sci 2(2):323343, Fu X, Zeng D, Huang Y, Zhang X, Ding X (2016) A weighted variational model for simultaneous reflectance and illumination estimation, In: IEEE Conference on Computer Vision and Pattern Recognition (CVPR) 2016:27822790, Guo X, Li Y, Ling H (2017) Lime: low-light image enhancement via illumination map estimation. In Rybaks theory, \({F^S}\) is defined as a local connection. [25] proposed a cascaded multistep binomial filtering histogram equalization (CMBFHE) method based on the POSHE algorithm. Neural Netw 61:85117, Shen D, Wu G, Suk H-I (2017) Deep learning in medical image analysis. Image Vision Comput 18(13):10051014, Cooper TJ, Baqai FA (2004) Analysis and extensions of the frankle-mccann retinex algorithm. Yang et al. [67] simplified the Retinex model for the first time based on partial differential equations (PDE). Due to the limitation of the paper length, we only elaborated the detailed algorithm of the above image quality evaluation indicators. There are many other image enhancement methods based on histogram modification that have not been listed, such as double histogram equalization, dynamic sub-histogram equalization, transform domain equalization and histogram clipping. In addition, due to the limitation of the paper length, here we only perform experimental evaluations of representative methods for each image enhancement type. Ma and Osher [81] introduced a total variation (TV) noise suppression model [82] to estimate the image illumination component. In summary, the image enhancement algorithm based on PCNN can be described as follows: Based on the capture characteristics and synchronous oscillation characteristics of PCNN, Li et al. Noise is an interference factor that must be considered in low-brightness image enhancement. [115] gave the transformation function as follows: Step 4: Normalize \({{{\mathcal {T}}}_{ij}}\) and round to the nearest integer, obtaining the enhanced image \({{{\mathcal {J}}}}\). Pattern Recognit Lett 5(2):129138, Morel J, Petro AB, Sbert C (2009) Fast implementation of color constancy algorithms. Then. The unsupervised algorithms do not require training samples or training labels, but directly model the data such as K-means [17], hierarchical clustering [18] and EM algorithm [19] etc. Image Enhancement is one of the most important Image processing techniques that is being used worldwide today. [90] proposed a new prior constraint, ie., local flatness. In this section, we first elaborates the quantitative evaluation index for the enhanced image. Compared with the logarithmic domain processing algorithm, PIE has the advantages of preserving details well and avoiding over-smooth reflection components. The algorithm flow of local histogram equalization. From Fig. \(\sigma\) represents kernel parameter, which can determine the enhancement effect of image. The "Enhance images in Microsoft Edge" setting is not available in release and insider channels, as of writing. In recent years, a large number of Retinex-like models have been also emerging. What is Image Processing? Meaning, Techniques, Segmentation & Important The quantitative measures we introduced here mainly involve absolute mean brightness error (AMBE) [153], structural similarity of image (SSIM) [154] and peak signal to noise ratio (PSNR), discrete entropy (DE) [155], enhancement (EME) [156] and lightness order error (LOE) [157]. [30] performed histogram equalization with maximum intensity coverage. J Opt Soc Am A Opt Image Sci Vision 32(11):21362147, Bertalmio M, Caselles V, Provenzi E (2009) Issues about retinex theory and contrast enhancement. [127]s method has the advantages of stable performance, fast speed, and outstanding effects for image enhancement, it also has many potential shortcomings. Representative methods include mean square error (MSE) and peak signal-to-noise ratio (PSNR). [87] and Tang et al. \({D_c}\) and \({D_t}\) are two adversarial discriminators, which involve the comparison of color and texture between the enhanced image and the high-quality image respectively. J Opt Soc Am A Opt Image Sci Vision 22(12):26132621, Marini D, Rizzi A (2000) A computational approach to color adaptation effects. For example, eliminating noise, revealing blurred details, and adjusting levels to highlight features of an image. Image enhancement is such a technique preprocessing where our goal is to suppress the noise while preserving the integrity of edges and the other detailed information (Gonzalez et al.,2014;. Here we conduct a review for GAN-like based image enhancement. J Test Meas Technol 19(3):304309, Li G-Y, Li H-G, Wu T-H (2005) Enhancement of image based on otsu and modified pcnn [j]. Finally, the lightness order error (LOE) is of great significance for describing the naturalness of the image, which is proposed to evaluate the quality of the image. The algorithm is described as follows: where \({P_k}(x)\) indicates the area of the k-th spot, L(x) is the input image. Sci Am 237(6):108128, Land EH, Mccann JJ (1971) Lightness and retinex theory. It is difficult to propose a method that has a good enhancement for all images. bf are the enhanced results of PIE [83], SRIE [85], LIME [86], Li et al. The evaluated image enhancement methods cover BIMEF [164], LIME [86], MSRCR [10], MF [163], SRIE [85], Dong [165], NPE [157] , DHECI [166], BPDHE [40], HE, Ying [168], WAHE [31], JED [167], Robust [87], LLNet [171], DeepUPE [169], GLADNet [170], MBLLEN [139] and Lv [141]. To solve this problem, this paper presents a new and effective method for image enlargement and enhancement based on adaptive inverse hyperbolic tangent . The main aim of image enhancementis to enhance thequality andvisual appearance of animage, or toprovide a better transformrepresentationfor future automated image processing. Qi, Y., Yang, Z., Sun, W. et al. IEEE Trans Image Process 24(12):49654977, Goldstein T, Osher S (2009) The split bregman method for l1-regularized problems. J Ambient Intell Hum Comput 10(10):40594070, Johnson JL, Padgett ML (1999) Pcnn models and applications. In particular, PIE is a processing algorithm based on linear space. J Optoelectron Laser 16(3):358362, Li G, Li H, Wu T (2005) The image enhancement based on modified pulse coupled neural network and genetic algorithm. In addition, (27) is usually simplified to \({F_{ij}}[n] = {S_{ij}}\). A Comprehensive Overview of Image Enhancement Techniques. Section2 introduces the image enhancement techniques based on unsupervised algorithms. In this theory, it is assumed that image lightness depends on illumination and reflectance. \(\phi\) denotes the time constants of inertial block and h represents the threshold of summarizing element. In their researches, the original image first is divided into lots of non-overlapping sub-blocks, and then histogram projection (HP) is performed respectively. In the work of Provenzi et al. It is worth noting that JED [167] and Robust [87] require a lot of computing resources, which can cause insufficient memory when processing large images. Acta Simulata Systematica Sinica 6:13701372, Qi C-L, Ma Y-D, Zhang Z-F (2006) Study on feedback pulse-coupled neural network model and its application. IEEE Trans Image Process 13(4):600612, Shannon CE (2001) A mathematical theory of communication. First of all, according to literature [153], the absolute mean brightness error(AMBE) is defined as: where MB(X) and MB(Y) represent the mean brightness of original image X and enhanced image Y, respectively. Image enhancement refers to the process of highlighting certain information of an image, as well as weakening or removing any unnecessary information according to specific needs. Motivated by this fact, Qi et al. \(V_L = 1\). [143]. Arch Comput Methods Eng 24(3):573588, Zhan K, Teng J, Shi J, Li Q, Wang M (2016) Feature-linking model for image enhancement. IEEE Trans Consum Electron 53(2):593600, Ibrahim H, Kong NSP (2007) Brightness preserving dynamic histogram equalization for image contrast enhancement. 21492159, Park J, Lee J, Yoo D, Kweon IS (2018) Distort-and-recover: Color enhancement using deep reinforcement learning, In: IEEE/CVF Conference on Computer Vision and Pattern Recognition 2018:59285936, Huang J, Zhu P, Geng M, Ran J, Zhou X, Xing C, Wan P, Ji X (2018) Range scaling global u-net for perceptual image enhancement on mobile devices, In: Proceedings of the European Conference on Computer Vision (ECCV) Workshops, Gharbi M, Chen J, Barron JT, Hasinoff SW, Durand F (2017) Deep bilateral learning for real-time image enhancement. IEEE 2018:15, Ying Z, Li G, Ren Y, Wang R, Wang W (2017) A new low-light image enhancement algorithm using camera response model. On the basis of Max-RGB [47], they considered the defect that it can only boost global illumination, and introduced the following initial estimation: According to Guo et al.s researches, \(\mathop B\limits ^ \wedge (x,y)\) in (26) guarantees that the recovery will not reach saturation. In addition, we also provide main quality evaluation methods for image enhancement. In addition to the above enhancement algorithm, Zhan et al. IEEE Trans Consum Electron 55(4):20722080, Singh K, Kapoor R (2014) Image enhancement via median-mean based sub-image-clipped histogram equalization. PubMedGoogle Scholar. Ma et al. [61] proposed a kernel-based Retinex model, which replace the two-dimensional spay process with a more efficient kernel. Finally, its Poisson equation is as follows: Horn first proposed to use an iterative strategy to solve it, but this is not the optimal method [56]. In order to preserve the brightness characteristics of the image and better enhance the local details of the image, many scholars have begun to devote themselves to the research of local histogram equalization(LHE) algorithms [4, 24,25,26]. In: IEEE International Conference on Consumer Electronics, Huang S, Cheng F, Chiu Y (2013) Efficient contrast enhancement using adaptive gamma correction with weighting distribution. \end{aligned}$$, $$\begin{aligned} {E_{ij}}[n]= & {} {e^{ - {\alpha _e}}}{E_{ij}}[n\mathrm{{ - 1}}]\mathrm{{ + }}{\mathrm{{V}}_E}{\mathrm{{Y}}_{ij}}[n] \end{aligned}$$, $$\begin{aligned} En{I_{ij}} = In(Bri) - (n - 1){\alpha _e} \end{aligned}$$, $$\begin{aligned} {M_{ijkl}} = {W_{ijkl}} = \frac{1}{{{{(i - k)}^2} + {{(j - l)}^2}}} \end{aligned}$$, $$\begin{aligned} {U_{ij}}\left( {{T_{ij}}} \right) > {E_{ij}}\left( {{T_{ij}}} \right) = {e^{{T_{ij}}}}{E_{ij}}\left( \mathrm{{0}} \right) \end{aligned}$$, $$\begin{aligned} {T_{ij}}\mathrm{{ = In}}\frac{{{U_{ij}}\left( {{T_{ij}}} \right) }}{{{E_{ij}}\left( 0 \right) }} \end{aligned}$$, $$\begin{aligned} {T_{ij}}\left( n \right) = {T_{ij}}\left( {n - 1} \right) + n{Y_{ij}}\left( n \right) \end{aligned}$$, $$\begin{aligned} {U_{ij}}\left[ n \right] = f{U_{ij}}\left[ {n - 1} \right] + {S_{ij}}\sum \limits _{kl} {{W_{ijkl}}{Y_{kl}}\left[ {n - 1} \right] } + {S_{ij}} \end{aligned}$$, $$\begin{aligned} {E_{ij}}\left[ n \right] = g{E_{ij}}\left[ {n - 1} \right] + h{Y_{ij}}\left[ {n - 1} \right] \end{aligned}$$, $$\begin{aligned} {Y_{ij}}\left[ n \right] = \left\{ \begin{array}{l} 1,{\quad } 1/(1 + \exp ( - \gamma ({U_{ij}}\left[ n \right] - {E_{ij}}\left[ n \right] ))) > 0.5\\ 0,{\quad } else \end{array} \right. Image Enhancement - an overview | ScienceDirect Topics The parameters n is the discrete iteration time of the neuron, \({V_F}\) and \({V_L}\) are the amplitudes of feedback input \({F_{ij}}\) and linking input \({L_{ij}}\) respectively, \({\alpha _f}\), \({\alpha _l}\) and \({\alpha _e}\) are the exponential decay coefficients of feedback input, linking input and dynamic threshold, respectively. Importance of Image Enhancement and CDF for Fault Assessment of Photovoltaic Module Using IR Thermal Image. Generative Adversarial Network (GAN) is a deep learning model and one of the most promising methods for unsupervised learning on complex distributions in recent years. The Retinex center/surround model was first proposed by Land [62], and its core idea is to estimate the illuminance component with a Gaussian kernel function. Rybak model was derived from the visual cortex system of guinea pigs, and its core structure is the iso-orientation domain model. Arch Computat Methods Eng 29, 583607 (2022). J R Stat Soc 39, Wang X, Chen L (2017) An effective histogram modification scheme for image contrast enhancement. Pattern Recognit Lett 104:1522, Zheng WT, Pu T, Cheng J, Zheng H (2012) Image contrast enhancement by contourlet transform and pcnn, In: International conference on audio, language and image processing 735739, Singh GMA (2014) Various image enhancement techniques-a critical review. Therefore, they proposed MSR-net to directly learn end-to-end enhanced mapping. From (35) we get the analytical solution of the time matrix \({T_{ij}}\), but \({T_{ij}}\) still cannot be obtained because (35) is an implicit function. In recently years, the double histogram equalization technology combines the ideas of image analysis, such as the local equalization of the background area, target area, texture area, and smooth area [37, 38]. \end{aligned}$$, $$\begin{aligned} {U_{ij}}\left[ n \right]= & {} f{U_{ij}}\left[ {n - 1} \right] + \left( {\sum \limits _{kl} {{M_{kl}}{Y_{kl}}\left[ {n - 1} \right] } + {S_{ij}}} \right) \nonumber \\\times & {} \left( {1 + \beta \left( {\sum \limits _{kl} {{M_{kl}}{Y_{kl}}\left[ {n - 1} \right] } - d} \right) } \right) \nonumber \\ {E_{ij}}\left[ n \right]= & {} g{E_{ij}}\left[ {n - 1} \right] + {V_E}{Y_{ij}}\left[ {n - 1} \right] \nonumber \\ {Y_{ij}}[n]= & {} \left\{ \begin{array}{l} 1,{\quad } if{\quad } {U_{ij}}[n] > {E_{ij}}[n] \\ 0,{\quad } else \end{array} \right. Various image enhancement algorithms were mentioned and underlying difficulties, limitations, merits and disadvantages were discussed in applying these techniques in the past two decades with three aspects: supervised algorithm, unsupervised algorithm and quality evaluation, respectively. The block diagram of the whole framework is shown in Fig. Kimmel et al. In 1974, Horn et al. Importance of Image Enhancement Techniques in Color Image Segmentation: A Comprehensive and Comparative Study Dibya Jyoti Bora Color image segmentation is a very emerging research topic in the area of color image analysis and pattern recognition. [26]. Neural Comput 2(3):293307, Reitboeck HJ (1989) A model for feature linking via correlated neural activity, Synergetics on Cognition, Parodi O, Combe P, Ducom JC (1996) Temporal coding in vision: coding by the spike arrival times leads to oscillations in the case of moving targets. Specifically, if the algorithm generates an output image, which introduces noise over homogeneous regions of the image, then although the output image looks unnatural, its corresponding EME value is very high [152]. Google Scholar, Dempster AP (1977) Maximum likelihood from incomplete data via the em algorithm. where \(p({x_i})\) is the probability of pixel intensity \({x_i}\), which is estimated from the normalized histogram. Histogram equalization achieves this function by effectively expanding the commonly used brightness. We will introduce in detail as follows: The whole framework of image enhancement methods. IEEE Trans Image Process 14(10):16471659, Chen Y-S, Wang Y-C, Kao M-H, Chuang Y-Y (2018) Deep photo enhancer: Unpaired learning for image enhancement from photographs with gans, In: Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, pp. Int J Comput Vision 83(1):101119, Land EH (1986) An alternative technique for the computation of the designator in the retinex theory of color vision. IEEE Trans on Image Process 27(6):28282841, Ren X, Yang W, Cheng W, Liu J (2020) Lr3m: robust low-light enhancement via low-rank regularized retinex model. [86] proposed a low-light image enhancement (LIME) method via illumination map estimation. (i, j)-th Rybak neuron-like element of the SNS. Chen et al. In the work of Qi et al., the linking item \({L_{12}}\) of RYNN1 and RYNN2 is set as follows: where \({\beta _{12}}\) is the local linking item between RYNN1 and RYNN2: \(\beta _{12} = max (I)/mean(I)\). The global item D(I) is used to control the final generated image and satisfies the GW (Gray world) assumption [76]. Comput Med Imaging Graph 24(2):5968, Kaur A, Singh C (2017) Contrast enhancement for cephalometric images using wavelet-based modified adaptive histogram equalization. Electron Lett 48(3):155157, Singh K, Vishwakarma DK, Walia GS, Kapoor R (2016) Contrast enhancement via texture region based histogram equalization. 2 in this paper. In addition, each sub-problem is transformed into a vectorized least squares regression, with closed-form solutions. https://doi.org/10.1007/s11831-021-09587-6, DOI: https://doi.org/10.1007/s11831-021-09587-6. [116] proposed a linking synaptic computation network (LSCN) and applied it to image enhancement. IEEE Trans Neural Netw 10(3):607614, Kuntimad G, Ranganath HS (1999) Perfect image segmentation using pulse coupled neural networks. IEEE Trans Image Process 18(9):19211935, Yun SH, Jin HK, Kim S (2011) Contrast enhancement using a weighted histogram equalization. Finally, the activation state of the neuron is determined, that is, the output \(Y_{ij}\) is 1 (activated) or 0 (not activated). \end{aligned}$$, https://doi.org/10.1007/s11831-021-09587-6. In image enhancement, in order to effectively eliminate the halo, scholars proposed the Retinex algorithm based on the variational framework, also known as the Retinex-like model. In this section, we elaborate the contents of the Retinex model as follows: According to Land et al. However, most of the existing deep learning based image enhancement methods rarely consider the characteristics of infrared imaging itself, and therefore hardly to balance the target enhancement and background suppression. The spiking cortical model (SCM) is a variant model of PCNN, which was proposed by Zhan et al. IEEE Trans Image Process A Publ IEEE Signal Process Soc 22(2):657667, Zhang R, Isola P, Efros AA, Shechtman E,Wang O (2018) The unreasonable effectiveness of deep features as a perceptual metric, In: IEEE/CVF Conference on Computer Vision & Pattern Recognition, Fu X, Zeng D, Huang Y, Liao Y, Ding X, Paisley J (2016) A fusion-based enhancing method for weakly illuminated images. Further, the pulse coupled neural network (PCNN) evolved from the Eckhorn model has been extensively and deeply studied by researchers. Signal Process 129:8296, Ying Z, Li G, Gao W (2017) A bio-inspired multi-exposure fusion framework for low-light image enhancement, CoRR abs/1711.00591 arXiv:1711.00591, Dong X, Pang YA, Wang G,Li W, Gao Y,Yang S, etal., A fast effecient algorithm for enhancement of low lighting video, J Inf Comput Sci 7(10) 20212030, Nakai K, Hoshi Y, Taguchi A, (2013) Color image contrast enhacement method based on differential intensity, saturation gray-levels histograms, In: International symposium on intelligent signal processing and communication systems. PSNR is calculated as follows: where peakval is the maximal variation of the input image data. In 2011, Yun et al. [87] proposed a Retinex model with a definite injected noise term, and for the first time tried to estimate the noise map based on the model. The characteristics of digital images justify the importance of an edge enhancement . However, histogram equalization is a method to adjust the image globally, which cannot effectively improve the local contrast, and the effect is very poor in some situations. The statistics of the number of papers with different image enhancement methods. For each training set, a model must be retrained, which is not universal. Look it up now! [119] proposed a heterogeneous Rybak neural network (HRYNN) as shown in Fig. Specifically, the greater the grayscale difference from the standard image, the more severe the image quality degradation. Proc IEEE Southeastcon95. Section3 elaborates the supervised algorithms for image enhancement. IEEE Computer Society Pr, Chang D, Wu W (1998) Image contrast enhancement based on a histogram transformation of local standard deviation. Thus, LOE is defined as follows: Due to the large amount of calculation of the LOE algorithm, in practical applications, the image is generally down-sampled with a sampling rate of \({{50} / {\min (m,n)}}\) to achieve the purpose of reducing the amount of calculation. Microsoft may have removed the Edge's controversial "Image Enhancement" feature following privacy concerns raised by users. \end{aligned}$$, $$\begin{aligned} \mathop {{\mathcal {R}}}\limits ^ \wedge = \arg \mathop {\min }\limits _R \left\{ {\left\| {\varDelta {{\mathcal {R}}} - {\delta _\varepsilon }(\varDelta {{\mathcal {L}}}(x,y)} \right\| _2^2} \right\} \end{aligned}$$, $$\begin{aligned} \mathop {{\mathcal {R}}}\limits ^ \wedge = \arg \mathop {\min }\limits _R \left\{ {{{\left\| {\varDelta {{\mathcal {R}}} - {\delta _\varepsilon }(\varDelta {{\mathcal {L}}}(x,y)} \right\| }_1}} \right\} \end{aligned}$$, $$\begin{aligned} {E_\omega }(I) = D(I) + {C_\omega }(I) \end{aligned}$$, $$\begin{aligned} D_{\alpha ,\beta }^\varepsilon (I): = \alpha \sum \limits _{x \in \Upsilon } {\left( {\frac{1}{2}\log \frac{1}{{2I(x)}} - \left( {\frac{1}{2} - I(x)} \right) } \right) } \nonumber \\ + \beta \sum \limits _{x \in \Upsilon } {\left( {{I_0}(x)\log \frac{{{I_0}(x)}}{{I(x)}} - \left( {{I_0}(x) - I(x)} \right) } \right) } \end{aligned}$$, $$\begin{aligned} C_\omega ^f(I): = \frac{1}{4}\sum \limits _{x \in \Upsilon } {\sum \limits _{y \in \Upsilon } {\omega (x,y)f\left( {\frac{{\min (I(x),I(y))}}{{\max (I(x),I(y))}}} \right) } } \end{aligned}$$, $$\begin{aligned} Minimize: F[{{\mathcal {B}}}]= & {} \iint _\varOmega \nonumber \\&\quad {\left( {{{\left| {\nabla {{\mathcal {B}}}} \right| }^2} + \alpha {{\left( {{{\mathcal {B}}} - {{\mathcal {L}}}} \right) }^2} + \beta {{\left| {\nabla \left( {{{\mathcal {B}}} - {{\mathcal {L}}}} \right) } \right| }^2}} \right) dxdy} \nonumber \\&\quad Subject to: {{\mathcal {B}}} \ge {{\mathcal {L}}} and \langle \nabla {{\mathcal {B}}},\mathop n\limits ^ \rightarrow \rangle = 0 on \partial \varOmega , \end{aligned}$$, \({{{\left| {\nabla {{\mathcal {B}}}} \right| }^2}}\), \({{{\left( {{{\mathcal {B}}} - {{\mathcal {L}}}} \right) }^2}}\), \({{{\left| {\nabla \left( {{{\mathcal {B}}} - {{\mathcal {L}}}} \right) } \right| }^2}}\), \({{{\left| {\nabla \left( {{\mathcal {R}}} \right) } \right| }^2}}\), $$\begin{aligned}&Minimize:F[{{\mathcal {B}}}] = \nonumber \\&\quad \iint _\varOmega {\frac{1}{2}\left( {\left\| {\nabla {{\mathcal {B}}}} \right\| _2^2 + t{{\left\| {\nabla {{\mathcal {R}}}} \right\| }_2}} \right) dxdy} \end{aligned}$$, \({\left\| {\nabla {{\mathcal {B}}}} \right\| _2^{}}\), $$\begin{aligned}&Minimize: F\left[ {{\mathcal {B}}} \right] = \nonumber \\&\quad \iint _\varOmega {\left( {\frac{\alpha }{2}\left\| {\nabla {{\mathcal {B}}}} \right\| _2^2 + {{\left\| {\nabla {{\mathcal {R}}}} \right\| }_2} + \frac{\beta }{2}{{\left\| {{{\mathcal {B}}} - {{\mathcal {R}}} - {{\mathcal {L}}}} \right\| }^2}} \right) dxdy} \end{aligned}$$, \({{\left\| {{{\mathcal {B}}} - {{\mathcal {R}}} - {{\mathcal {L}}}} \right\| }^2}\), $$\begin{aligned} \mathop B\limits ^ \wedge (x,y) \leftarrow \mathop {\max }\limits _{c \in \left\{ {R,G,B} \right\} } {L^c}(x,y) \end{aligned}$$, $$\begin{aligned} {F_{ij}}[n]= & {} {e^{ - {\alpha _f}}}{F_{ij}}[n\mathrm{{ - 1}}] + {V_F}\sum \limits _{kl} {{M_{ilkl}}{Y_{kl}}[n - 1]} + {S_{ij}} \end{aligned}$$, $$\begin{aligned} {L_{ij}}[n]= & {} {e^{ - {\alpha _l}}}{L_{ij}}[n\mathrm{{ - 1}}]+{V_L}\sum \limits _{kl} {{W_{ilkl}}{Y_{kl}}[n - 1]} \end{aligned}$$, $$\begin{aligned} {U_{ij}}\left[ n \right]= & {} {F_{ij}}\left[ n \right] \left( {1 + \beta {L_{ij}}\left[ n \right] } \right) \end{aligned}$$, $$\begin{aligned} {Y_{ij}}[n]= & {} \left\{ \begin{array}{l} 1,{\quad } if{\quad } {U_{ij}}[n] > {E_{ij}}[n]\\ 0,{\quad } else \end{array} \right. In 2019, Nie et al. For the exterior scenes in the paper, lots of images of blue sky and sea were selected. Generally, image enhancement algorithms based on time matrix can be summarized as follows: Step 1: Normalize the image intensity I by: where min(I) is the minimum value of I and max(I) is the maximum value of I. Comput Eng Appl 52(8):163168, Lei L, Xi F, Chen S (2019) Finger-vein image enhancement based on pulse coupled neural network. U(x,y) is step functions, and \(\oplus\) represent the exclusive OR operation. It is usually described as follows: where l(x,y), c(x,y) and s(x,y) denote brightness comparison part, contrast comparison part and structure comparison part of image, respectively. In addition, the Mach zone phenomenon is explained through experiments and the rationality of the method is established. 68496857, Wang W, Wei C, Yang W, Liu J, Gladnet: Low-light enhancement network with global awareness, In: 2018 13th IEEE International Conference on Automatic Face & Gesture Recognition (FG 2018), IEEE, 2018, pp. It is often affected by various factors such as experimental conditions, the observers knowledge background, mood, motivation, and fatigue. 844848, Yang X, Liu T, Li X (2016) Study on image enhancement algorithm merged wavelet transform and improved pcnn [j]. Edge Enhancement in Indoor Digital Images | Oriental Journal of Shen et al. It consists of a leakage integrator, linking input and feedback input as follows. In this section, we will conduct a comprehensive and systematic experimental evaluation of image enhancement methods. Cooper et al. The receptive field \({F^S}(x,y)\) has been proved to have multi-scale adjustable characteristic, which is sensitive to the detailed texture information of the image. Neurocomputing 4(12):93102, Yang J, Wang G, Zhang Z, Lin S, Lu T, Jiang L (1996) The edge-detecting system of image based on the artificial receptive field implemented by the bacteriorhodopsin films. Visualize the Future, IEEE, pp 3743, Lindblad T, Kinser JM, Taylor J (2005) Image processing using pulse-coupled neural networks.
Afternoon Tea Willingham Auctions, Articles I