图像处理中的软计算与机器学习进展

需积分: 1 109 浏览量更新于2024-07-18 收藏 24.75MB PDF 举报

Advances in Soft Computing and Machine Learning in Image Processing 本书《Advances in Soft Computing and Machine Learning in Image Processing》是关于软计算和机器学习在图像处理领域的最新进展。该书汇集了关于机器学习技术与图像处理方法结合的最佳章节。这些主题不仅在科学社区中非常重要，也在工程和医学等领域中具有广泛的应用前景。在图像处理和计算机视觉领域，软计算和机器学习技术的应用非常广泛。本书中每章节都讨论了图像处理和计算机视觉领域中的重要问题，并且展示了软计算和机器学习技术在这些问题中的应用。软计算是一种基于模糊逻辑、神经网络和遗传算法等技术的计算方法。它可以处理不确定性和模糊性问题，具有很强的鲁棒性和泛化能力。在图像处理领域，软计算技术可以应用于图像增强、图像分割、图像识别等领域。机器学习是指利用机器学习算法来自动学习和改进图像处理方法的技术。机器学习算法可以根据图像特征来分类、聚类和回归图像。这些算法可以应用于图像识别、图像分类、图像检索等领域。本书还讨论了软计算和机器学习技术在工程和医学等领域中的应用。例如，在医学成像领域，软计算和机器学习技术可以应用于图像分割、图像识别和图像分析等方面。本书《Advances in Soft Computing and Machine Learning in Image Processing》是关于软计算和机器学习在图像处理领域的最新进展的综述。该书涵盖了软计算和机器学习技术在图像处理领域中的应用，并讨论了这些技术在工程和医学等领域中的前景。知识点： 1. 软计算技术在图像处理领域中的应用 2. 机器学习技术在图像处理领域中的应用 3. 软计算和机器学习技术在图像处理领域中的结合 4. 软计算技术在图像增强、图像分割和图像识别等领域中的应用 5. 机器学习算法在图像分类、图像检索和图像识别等领域中的应用 6. 软计算和机器学习技术在工程和医学等领域中的应用前景了解更多关于软计算和机器学习在图像处理领域中的应用，可以查看相关的研究论文和技术报告。

1 Colors Models and Spaces

Color is one of the most important aspects in vision, since it is used generally to

discriminate and recognize information. From the Physics view, the color is caused

by the reﬂection of a portion of light in an object. Basically, it is just an electro-

magnetic wave in a range of frequencies. The longest wavelength produces the red

and the shortest blue-violet (Fig. 1).

The part of the electromagnetic spectrum analyzed is the one called the visible

spectrum, which contains all the possible colors. Other colors existing outside the

visible spectrum range are not contemplated in this section (Infrared and ultraviolet)

[15].

Since representing the colors as a speciﬁc wave frequency with a certain

amplitude and longitude is not trivial or natural to human beings, it created systems

that help to describe the colors in a more natural and ordere d way. The descriptions

systems are known as color models.

A color model is actually a set of equations and/or rules used to calculate a given

color and are capable of describing a wide range of tones from the visible spectrum

of light.

A color space is deﬁned as all the possible tones generated by a color model; in

this work the colors are managed from the color space perspective.

1.1 Additive Color Spaces

Additive color spaces are those set of colors that are created by mixing primary

colors. The most used primary colors are the red, blue, and green. Originally, these

colors were formed by adding two or more rays of light into an obscure surface. The

created color space using this technique is named RGB [23].

Fig. 1 Visible light spectrum. Wavelengths are in meters

4 E. Chavolla et al.

A variant of the additive color spaces is the subtractive color space, which

instead of adding rays of light, it blocks or subtracts certa in frequencies from a ray

of light containing the complete frequency range from the visible light spectrum.

This section will use only the additive color space in form of the RGB model,

since it is the most frequent color space used in computer images.

RGB This model is most widely used in computer graphics. Many works use this

model as a base for color ﬁltering, segmentation, and analysis in image processing.

The ease of use and intuitive model for color creation and manipulation makes it an

ideal choice for drawing and coloring (Fig. 2).

Unfortunately, it has some disadvantages when it is used in image processing.

The model produces a nonlinear and discontinues space, which makes the changes

in color hue hard to follow due to discontinuities. Another issue found in RGB

model is the fact that the color hue is easily affected by illumination changes,

making the color tracking and analysis a nontrivial task.

Despite the issues described for the RGB model, the produced space is preferred

in most of the computational tasks. One of the reasons is that most of the computer

hardware and software is developed under this model, so it can be used directly

without additional work.

Normalized RGB This model is a variation from the RGB, and basically is

created under the premise of protecting the color model from the illumination

changes. The main idea behind this variation is that the colors are formed using a

certain proportion of three primary colors from the model, not a deﬁned amount of

each one. So the sum of all the colors proportions must equal a 100%. The process

of normalization is straightforward and can be easily computed (Eqs. 1–3). Basi-

cally, each of the equations used to transform RGB to normalized RGB obtains the

proportion of each color. Another advantage of this model is that any color can be

described just using two colors instead of three. This is possible due to the fact that

the third color will always be the difference between one and the addition of the

other two colors:

Fig. 2 RGB color space

(Wikimedia commons RGB)

Color Spaces Advantages and Disadvantages in Image … 5

r =

R + G + B

ð1Þ

g =

R + G + B

ð2Þ

b =

R + G + B

. ð3Þ

This variant certainly reduces the negative effect of the illumination changes like

shadows or shines, but in exchange it reduces object detection capabi lity. This is

due to the loss of contrast that the same illumination provides [5].

This model is used in machine learning with some degree of success, and it will

part of the comparison performed in this work

1.2 Perceptual Color Spaces

Perceptual color spaces are created from those models that treat the color in a more

human intuitive form. In order to achieve this objective, any color is represented by

a speciﬁc tone or hue, a level of saturation for the hue, and the amount of light

available or illumination.

Inside this category, it can be found Hue Saturation Value model (HSV), Hue

Saturation Lightness (HSL), and Hue Saturation Intensity (HSI). All of these color

models have a similar description about the color hue but differ in the saturation and

illumination deﬁnition.

HSL This color model can be computed from the RGB standard model by using a

set of equations (Eqs. 4–6). The major advantage of using this model is the fact that

it presents immunity to illumination changes, since the illumination is enclosed in

the lightness component of the model. The other feature of this model can be found

in the color hue changes; they are continuous and linear. The hue component is

usually expressed in angle terms, from 0 to 359°, since every tone has a previous

and next tone that follows a cyclic pattern and can be visualized in a circular

diagram. The last feature of the HSL model is the geometric representation which

can be expressed as a bi-conic ﬁ gure or a cylinder (Fig. 3), which allows to

manipulate easily each color hue by changing its illumination and saturation [7]:

60 *

G − B

max R, G, BðÞ− min R, G, BðÞ



R = max R, G, BðÞ

60 *

2+ B − RðÞ

max R, G, BðÞ− min R, G, BðÞ



G = max R, G, BðÞ

60 *

4+ R − GðÞ

max R, G, BðÞ− min R, G, BðÞ



B = max R, G, BðÞ

ð4Þ

6 E. Chavolla et al.

剩余710页未读，继续阅读

wang1062807258

粉丝: 13
资源: 272

图像处理中的软计算与机器学习进展

Advances in Soft Computing and Machine Learning in Image Processing 无水印pdf

Soft Computing in Machine Learning(PDF)

Advances in Low-Level Color Image Processing

Advances in Soft Computing and Machine Learning in Image Processing (2018)

Recent Advances on Soft Computing and Data Mining

Advances in Machine Learning and Signal Processing

Advances in Neural Computation, Machine Learning, and Cognitive Research

Recent Advances in Soft Computing(2017)[Radek_Matoušek_(eds.)].pdf

Advances in Machine Learning and Data Analysis

Advances in Cloud-Scale Machine Learning for Cyber-Defense - Russinovich 2017 (exp-t11-advances-in-cloud-scale-machine-learning-for-cyber-defense)-计算机科学

最新资源