启发式算法，代码和参考文献_启发式算法是什么资源-CSDN文库

共24个文件

pdf：18个

caj：3个

m：2个

蝙蝠算法

需积分: 12 112 浏览量 2022-10-06 19:55:06 上传评论收藏 48.38MB RAR 举报

资源详情

资源评论

资源推荐

收起资源包目录

参考文献.rar （24个子文件）

参考文献

关于磷虾群优化算法在约束优化问题的研究与应用_杨文哲.caj 5.08MB

s11042-020-10139-6.pdf 2.14MB

s10462-017-9559-1.pdf 936KB

Comparative_Study_of_Krill_Herd_Firefly_and_Cuckoo.pdf 1.07MB

几种改进遗传算法的性能比较_刘刚.pdf 388KB

1-s2.0-S0957417421008927-main.pdf 2.75MB

1-s2.0-S0957417420310848-main.pdf 1.07MB

基本蚁群-AS

AC_TSP.m 3KB

鳞虾群算法的改进及其在结构可靠性分析中的应用_程立翔.caj 5.02MB

1-s2.0-S095070512030736X-main.pdf 15.76MB

1-s2.0-S0305054814002652-main.pdf 478KB

多时间窗车辆路径问题的智能水滴算法_李珍萍.pdf 572KB

s00521-020-04832-8.pdf 1.35MB

1-s2.0-S1007570412002171-main.pdf 1.49MB

蚁群.py 9KB

IWD.pdf 1.07MB

基于改进蚁群算法的电动汽车充电调度研究_蔡志民.caj 10.01MB

Elephant_Herding_Optimization.pdf 251KB

原文.pdf 742KB

势场算法

typical_algorithm.m 3KB

1-s2.0-S0957417420310848-main (1).pdf 1.07MB

8134674.pdf 1.65MB

1-s2.0-S2214914718300333-main.pdf 3.46MB

eusuff2006.pdf 668KB

Knowledge-Based Systems 212 (2021) 106607

Contents lists available at ScienceDirect

Knowledge-Based Systems

journal homepage: www.elsevier.com/locate/knosys

Moth Swarm Algorithm for Image Contrast Enhancement

Alberto Luque-Chang

∗

, Erik Cuevas

, Marco Pérez-Cisneros

, Fernando Fausto

Arturo Valdivia-González

, Ram Sarkar

Departamento de Electrónica, Universidad de Guadalajara, CUCEI Av. Revolución 1500, 44430, Guadalajara, Mexico

Department of Computer Science and Engineering, Jadavpur University, Kolkata, 700032, India

a r t i c l e i n f o

Article history:

Received 24 June 2020

Received in revised form 12 October 2020

Accepted 8 November 2020

Available online 26 November 2020

Keywords:

Image enhancement

Moth Swarm Algorithm

Global optimization

a b s t r a c t

Image Contrast Enhancement (ICE) is a crucial step in several image processing and computer vision

applications. Its main objective is to improve the quality of the visual information contained in the

processed images. The presence of noise and small sets of pixels in images are not only irrelevant

for their visualization. It also negatively affects the improvement process of ICE schemes since the

inclusion of irrelevant information avoids the appropriate distribution of significant pixel intensities

in the enhanced image. As a consequence of this effect, most of the proposed ICE methods present

different associated problems such as the production of undesirable artifacts, noise amplification, over

saturation and bad human visual perception. In this paper, an Image Contrast Enhancement (ICE)

method for grayscale and color images is presented. The proposed approach has the propriety of

eliminating noisy and irrelevant information in order to improve the distribution capacity of significant

pixel intensities in the enhanced image. Our method eliminates multiple groups of a very small number

of pixels that, according to their characteristics, do not represents any object or important detail of the

image. This process is done by the Mean-shift algorithm, which is used to replace such sets of irrelevant

pixels in the original histogram by significant pixel densities represented by local maxima. Then, the

Moth Swarm Algorithm (MSA) is used to redistribute the pixel intensities of the reduced histogram

so that the value from Kullback–Leibler entropy (KL-entropy) has been maximized. The proposed

approach has been tested considering different public datasets commonly used in the literature. Its

results are also compared with those produced by other well-known ICE techniques. Evaluation of the

experimental results demonstrates that the proposed approach highlights the important details of the

image also improving its human visual appearance.

1. Introduction

A Knowledge-Based Systems (KBS) [1] is an area of artificial

intelligence that refers to the extraction and representation of the

knowledge in engineering. The information extraction through

optimization principles in computer vision lies at the heart of

KBS. Most of the KBS problems applied to image analysis can

be reduced into optimization processes. Under this methodology,

information contained in the image is evaluated considering a

knowledge base represented by a set of important quality char-

acteristics. Then, by using a search strategy, the best image is

detected. This image corresponds to the solution of the KBS

problem.

∗

Corresponding author.

E-mail addresses: alberto.lchang@academicos.udg.mx (A. Luque-Chang),

erik.cuevas@cucei.udg.mx (E. Cuevas), marco.perez@cucei.udg.mx

(M. Pérez-Cisneros), abraham.fausto@academicos.udg.mx (F. Fausto),

arturo.valdivia@academicos.udg.mx (A. Valdivia-González), ramjucse@gmail.com

(R. Sarkar).

Image enhancement (IE) is a computer vision task that can be

approached as a KBS problem. It has attracted the attention of

the computer vision community due to its multiple applications

in areas such as medicine, security, transportation, etc [2–6]. IE is

the process of improving the visual information contained in an

image, increasing the difference among features of its different

objects. The main objective is to improve the interpretability

of the information present in an image for human viewers or

make the enhanced image more suitable for further processing

steps in any automatic computer vision system [6,7]. In general,

IE methods modify pixel values through the histogram equal-

ization, quadratic transformation, or fuzzy logic operation[2–4].

Among these techniques, histogram equalization (HE) [5,7–10] is

the most used, simple and effective for image enhancement. HE

considers the statistical features of pixels. In its operation, pixels

relatively concentrated in positions of the histogram are redis-

tributed over its whole scale. During this process, each existent

intensity value A of the original image is mapped into another

value B in the processed image without matter the number of

pixels corresponding to A in the original image. Therefore, these

https://doi.org/10.1016/j.knosys.2020.106607

A. Luque-Chang, E. Cuevas, M. Pérez-Cisneros et al. Knowledge-Based Systems 212 (2021) 106607

schemes present an inadequate redistribution of the pixel data

under the presence of noise or irrelevant small sets of pixels.

As a consequence, these methods produce enhanced images with

different problems such as the generation of undesirable artifacts

and noise amplification [11–13].

In the last years, the problem of ICE has been approached

through metaheuristic techniques as an alternative to design ICE

schemes. Different from HE techniques, metaheuristic methods

face the contrast enhancement problem from an optimization

perspective. Therefore, under these methods, some aspects of the

contrast enhancement process are related to an objective func-

tion whose optimal solution represents a good quality enhanced

image. In general, contrast enhancement algorithms based in

metaheuristic schemes have demonstrated to produce better re-

sults than those delivered by HE in terms of pixel intensity

redistribution [3]. Some examples of these approaches include the

use of Artificial Bee Colony (ABC) [14], Genetic Algorithms (GA)

[12] and Particle Swarm Optimization (PSO) [15], Social Spider

Optimization (SSO) [11,13], Cuckoo Search Algorithm (CS) [16],

Firefly Algorithm (FFA) [17], Gravitational Search Algorithm (GSA)

[18], Differential Evolution (DE) [19], among others. Although

these methods have produced interesting image enhancement

results, they present several flaws such as premature convergence

to local optima and inability to maintain population diversity.

In these methods, the redistribution of pixel data is evaluated

through objective functions [12] that consider mainly hetero-

geneity. The more different pixel intensities are contained in

the enhanced image, the more quality is the distribution. Under

such conditions, these schemes present a limited effect of con-

trast enhancement in the presence of noise and irrelevant small

set of pixels. This effect cannot be improved unless irrelevant

visual information could be removed before the redistribution.

Recently, new hybrid approaches that combine metaheuristic

techniques with other computational schemes have been intro-

duced in the literature for image contrast enhancement proposes.

The objective of such methods is to increase the capacities of

the metaheuristic approach through the incorporation of specific

computational elements that provide additional information on

the contrast enhancement process. Some representative exam-

ples include the algorithm Gamma-PSO proposed in [20], where

the optimal values of a Gamma correction are determined by the

PSO technique in order to enhance the contrast of a previously

transformed image with a wavelet transform. Another interesting

work is the Fuzzy-IPSO method reported in [21], where it is

introduced an approach that combines fuzzy logic and PSO for en-

hancing the contrast of an image. Under this method, information

of the histogram is obtained by a fuzzy system. With these data,

the histogram is adjusted by the use of some weights computed

through an improved version of the PSO algorithm.

The Moth Swarm Algorithm (MSA) [22] is a metaheuristic

scheme inspired by the orientation of moths towards the moon-

light. Different from other metaheuristic schemes, MSA considers

three specific sub-populations. Therefore, according to the sub-

population, each individual is conducted by using a different evo-

lutionary operation, which represents a particular behavior of the

moth insect. The incorporation of these operators in the search

strategy reduces critical problems present in several metaheuris-

tic algorithms, such as improper exploration-exploitation balance

and premature convergence. Due to its interesting characteristics,

MSA has been extensively applied in many complex engineering

problems such as image processing [23], power systems [22],

energy conversion [24], etc.

The presence of noise and small sets of pixel intensities in

images often refers to the existence of ineffective/useless charac-

teristics that maintain important image details hidden. Such ele-

ments correspond to sets of pixels that are too small to represent

objects or visible details in the image. This information contained

in the image is not only adverse for human visualization, but

also negatively influence the process of image enhancement. The

elimination of this information represents a necessary method

to improve the redistribution of the pixel data in the enhanced

image. Therefore, relevant pixel intensities will be correctly high-

lighted in the redistribution since the space maintained for the

eliminated pixel intensities can be reused. The Mean-shift al-

gorithm [25] is a non-parametric technique commonly used to

replace a characteristic x scarcely existent in the feature space

by an element x

∗

densely contained within its neighborhood

(local maximum). Under this scheme, starting at point x, a new

location is computed in the direction of the largest increase of the

feature space until the local maximum x

∗

has been reached. The

simplicity and interesting properties of Mean-shift have extended

its use in several domains such as regression [25] and clustering

[26].

On the other hand, the Kullback–Leibler entropy (KL-entropy)

[27,28], introduced by Kullback and Leibler, allows objectively

evaluating the divergence between two different probability dis-

tributions. In general terms, the KL-entropy is used to assess how

one probability distribution is different from a second probability

considered as a reference. Its minimal value represents the com-

plete dissimilarity between both distributions, while its maximal

value corresponds to its best resemblance. Applications of the KL-

entropy include the evaluation of the randomness in continuous

time-series, the characterization of the relative (Shannon) en-

tropy in information systems and the measurement of contained

information when comparing different statistical models of infer-

ence. Different from other entropy formulations, the KL-entropy

allows us to obtain no only the best possible data distribution

in terms of the information content but also the distribution

with the best similitude with regard to the original image data.

Under such conditions, the results delivered by the use of KL-

entropy involve images with enhanced contrast, also improving

their human visual appearance (without distortion regarding the

original information).

In this paper, an Image Contrast Enhancement (ICE) algorithm

for images is introduced. Different from other approaches, the

proposed ICE method considers the previous elimination of irrel-

evant sets of pixel intensities. The objective of this elimination

is to reduce the number of histogram features in order to use

the whole histogram range in the redistribution produced by

the enhancement process. During this elimination process, the

Mean-shift method is employed to delete unimportant small

sets of pixels through the replacement of sporadic intensities by

abundant features. Once obtained, the reduced histogram, the

contrast improvement process, is performed. In this process, the

Moth Swarm Algorithm (MSA) is adopted to redistribute the pixel

intensities of the reduced histogram in the complete range so that

the value of the Kullback–Leibler Entropy (KL-entropy) between

a candidate distribution and the original information has been

maximized. This paper presents three important contributions:

(I) The use of an elimination process to reduce the number

of histogram features in order to improve the pixel intensity

redistribution. Without consideration of this process, the level

of contrast enhancement reached is very limited. According to

our best knowledge, no comprehensive study has considered

the elimination process of irrelevant information to improve the

redistribution process within an ICE scheme. (II) The incorpo-

ration of the KL-entropy for the redistribution process. Though

its inclusion, images with a better human visual appearance are

produced. (III) The use of the Moth Swarm Algorithm (MSA) to

find the best redistribution that represents the best-enhanced im-

age. The search characteristics of this scheme allow exploring the

solution space without presenting the associate problems of other

A. Luque-Chang, E. Cuevas, M. Pérez-Cisneros et al. Knowledge-Based Systems 212 (2021) 106607

metaheuristic schemes. The performance of the proposed scheme

has been tested considering a number of representative datasets

commonly found in the literature. Its results are also compared

with those produced by other well-known similar techniques.

Experimental results suggest that the proposed method has a

better performance in comparison with other schemes in terms

of several performance indexes.

The rest of the paper is organized as follows: In Section 2, a

brief explanation of the Mean-shift algorithm is given; in Sec-

tion 3, the main characteristics of the ICE problem are illustrated;

in Section 4, the most important elements about the MSA are

presented; in Section 5, the proposed approach is presented;

Section 6 exhibits the experimental results; finally, in Section 7,

conclusions are drawn.

2. Mean-shift algorithm

The Mean-shift method [25] is a non-parametric scheme com-

monly used to reduce unimportant information contained in a

dataset. Under this algorithm, a characteristic x scarcely present

in the feature space is replaced by other x

∗

densely contained

within its neighborhood (local maximum). The operation of the

Mean-shift starts at a random point x within the density fea-

ture map. Then, a new location is computed in the direction of

the largest increase of the density feature space until the local

density maximum x

∗

has been reached. Two processes integrate

the Mean-shift scheme. First, a Kernel function is applied to

estimate the probability density function (PDF) of the data. Then,

an optimization step based on gradient ascent is conducted to

find the local maxima (density attractors) over the PDF [29].

2.1. Probability density function estimation

In the mean-shift algorithm, the probability density function

is estimated defining some non-negative, symmetric kernel func-

tion K, such that:

K(x) ≥ 0; K

(

−x

)

= K (x);



K(x)dx = 1 (1)

Assuming a set of n features x

, x

, . . . , x

, the density func-

tion

f (x), can be estimated in terms of K as follows:

f (x) =



i=1



x − x



, (2)

where h represents the influence area of K.

2.2. Density attractors

The main idea behind the estimation of a PDF for a given

dataset is to find characteristic points x

∗

which better represent

the distribution of the modeled data points. These points are

represented by the local maxima present in the estimated PDF

f (x). Such points can be found by applying a gradient-based

optimization method. In this scheme, a random point x is first

defined. Then, a new location is computed in the direction of

the density gradient ∇

f (x) until the local maximum x

∗

has been

reached. This update rule is formulated as follows:

t+1

= x

+ δ · ∇





, (3)

where t denotes the current iteration number, while δ represents

the step size. Under this formulation, the density gradient at any

point x is calculated by computing the derivative of the PDF as

follows:

∇

f (x) =

∂

∂x

f (x) =



i=1

∂

∂x



x − x



(4)

Fig. 1. Illustration of the mean-shift process. (a) dataset example, (b) estimated

PDF

f (x) from Eq. (2) and (c) replacing of x by x

∗

as a consequence of the

density gradient operation.

Fig. 1 is an illustration of the mean-shift process. Considering

that Fig. 1(a) represents the dataset, Fig. 1(b) shows the estimated

PDF

f (x) from Eq. (2). Then, starting at point x, a new location is

computed in the direction of the largest increase of the feature

space until the local maximum x

∗

has been reached. Therefore,

the feature x of low density is replaced by x

∗

which represents a

neighbor representative feature.

3. Image contrast enhancement through optimization

An optimization-based contrast enhancement approach re-

lates some aspects of the contrast enhancement process to an

objective function. Under such conditions, obtaining the minimal

or maximal value of this objective function, it is found the image

with the best possible contrast enhancement characteristics in

terms of the elements used for the construction of the objective

function. Under these schemes, it is coded a set of alternative

histogram distributions as decision variables to determine hy-

pothetical solutions to the contrast enhancement problem. Each

histogram is represented by a vector of 256 dimensions. An

objective function evaluates the quality of the enhancement pro-

duced by each hypothetical histogram. Therefore, conducted by

the values of the objective function, the set of encoded his-

tograms is operated through a particular metaheuristic approach

so that the solutions (histograms) are iteratively improved in

each generation of the optimization process. In general, a contrast

enhancement task involves the maximization of the objective

function, the lower the index, the better is the improved result.

Fig. 2 illustrates the enhancement process under an optimiza-

tion approach. In the figure, a solution A, is modified by the

operators of the optimization approach. As a consequence of this

modification, a new solution B is produced. The new solution B

A. Luque-Chang, E. Cuevas, M. Pérez-Cisneros et al. Knowledge-Based Systems 212 (2021) 106607

Fig. 2. Illustration of the enhancement process under an optimization approach.

presents a better enhancement index according to its evaluation

of the objective function.

4. Moth swarm algorithm

The MSA [22] is a recent swarm algorithm designed to solve

global optimization problems. In MSA, the swarm population is

divided into three groups: Pathfinders, Prospectors and Onlook-

ers. The following subsections are summarized the complete MSA

algorithm. For more details, refer to [22].

4.1. Initialization

In the MSA approach, the population involves a set of p indi-

vidual moths M =



, m

, . . . m



, where m



i,1

, m

i,2

, . . . ,

i,n



represents a candidate solution. During the initialization,

the search agents assume random positions according to the

following model:

= rand

(

0, 1

)



max

− m

min



+ m

min

(5)

∀i ∈

{

1, 2, . . . , p

}

, j ∈ {1, 2, . . . , n}

where, m

max

and m

min

are the upper and lower limits of the search

space.

4.2. Recognition phase

In this phase, two operators are applied to each candidate

solution: crossover and Lévy perturbation [30]. The crossover

point is defined through the following formulation:





a=1



− m





a=1

(6)

with p

is the number of pathfinder moths. Under such condi-

tions, the new solution is computed as follows:



j=1

(7)

4.2.1. Lévy flights

The MSA uses the Lévy perturbations [30] to generate random

steps. Therefore, to generate a Lévy L

random sample the Man-

tegna’s algorithm [31] is used. This scheme can be described as

follows:

∼ scale ⊕ L

evy

(

)

∼ 0.01

, (8)

where α ∈

[

0, 2

]

, scale is the dispersion size, ⊕ is the entrywise

multiplication, v = N(0, µ

) and z = N(0, µ

) are two normal

distributions.

4.2.2. Lévy Mutation

For nc operations (nc ∈ cp), the algorithm creates a sub-trial

vector

−→

= [x

, x

, . . . , x

] as follows:

−→

= L

∗



−→

, −,

−→



+ L

∗



−→

, −,

−→



(9)

∀r

= r

= p ∈

{

1, 2, . . . , np

}

where L

and L

are two independent Lévy random samples [31].

4.2.3. Adaptive crossover

Each pathfinder updates its position combining crossover op-

erations and mutated variables. Therefore, this trail/mixed solu-

tion Tms

for the element i is described as follows:

Tms







−→

if j ∈ cp

−→

if j /∈ cp

(10)

4.2.4. Selection strategy

Once the adaptive crossover operator is complete, the fitness

value of the complete trail solution is calculated and compared

with its corresponding host solution. Then, a set of np solutions

are selected following the roulette scheme [22].

4.3. Transversal flight

The group of elements with the best luminescence intensities

is considered as prospectors in the next iteration. On the course

A. Luque-Chang, E. Cuevas, M. Pérez-Cisneros et al. Knowledge-Based Systems 212 (2021) 106607

Fig. 3. The flowchart of the MSA algorithm.

of the optimization process, the prospectors number pf decreases

as follows:

pf = round



(

p − p

)



1 −



, (11)

where i

is the current iteration number.

4.4. Celestial navigation

During the optimization process, the decreasing prospectors

number increases the onlookers number (o

= p − p

− p

). This

may lead to a fast increment on the convergence rate. The moths

with the lower luminescent sources in the swarm are considered

onlookers. In this phase, onlookers search exhaustively near to

prospector elements. The onlookers are divided into two equal

parts.

4.4.1. GaussIan walks

In this phase, with the objective of exploiting promising areas

on the search space, a stochastic Gaussian distribution is used to

perturb original positions. Therefore, the new onlooker positions

评论收藏

内容反馈

ustc懒苗

粉丝: 1026
资源: 4

启发式算法，代码和参考文献

评论0

最新资源

启发式算法，代码和参考文献

评论0

启发式算法,启发式算法有哪些,matlab

【老生谈算法】CDS启发式算法及Matlab程序.docx

packing_java_超启发式算法_

启发式算法阅读材料

元启发式算法文档

NSGA-III算法实现文献参考

深度学习参考文献1

毫米波雷达心跳呼吸信号提取算法及参考文献

数学建模算法全收录参考文献

启发式算法-数学建模-代码

数学建模启发式算法

生产排程之启发式算法

高级算法实验代码及报告--启发式算法

matlab写启发式算法代码-CS170_Project1:CS170_Project1

停机位分配启发式算法源码

论文研究 - 求解机器调度问题的Job Shop启发式算法

Novel Bat Algorithm (NBA) 元启发式算法 - MATLAB代码

超启发式算法介绍大全

四种经典启发式算法求解TSP问题

数学建模培训 MATLAB与数学建模 启发式算法简介及其在数学模型中的应用 共147页.pdf

完整车牌号识别程序，可以识别车牌和颜色，可以集成到项目中 支持win7+

ChatGPT教程（终极版）最全整理

博客中Kmeans以及FCM算法数据（免积分）

神经网络回归预测--气温数据集

hugging face的models-openai-clip-vit-large-patch14文件夹

XGBoost+LightGBM+LSTM-光伏发电量预测

Mathwork+Matlab+编程手册

时间序列预测模型实战案例(Xgboost)(Python)(机器学习)包括时间序列预测和时间序列分类，点击即可运行！

中文短信数据集-带标签

最新资源

数学建模培训 MATLAB与数学建模启发式算法简介及其在数学模型中的应用共147页.pdf

完整车牌号识别程序，可以识别车牌和颜色，可以集成到项目中支持win7+