2009年全面图像处理指南：学术出版社经典

5星 · 超过95%的资源需积分: 9 152 浏览量更新于2024-07-22 收藏 33.54MB PDF 举报

《图像处理基础指南2009ap》是由学术出版社（AcademicPress）旗下的艾利斯维尔（Elsevier）发行的一本权威参考书籍。该书以其全面且深入的内容，在图像处理领域享有盛誉。作为一部2009年的出版物，它反映了当时最先进的图像处理技术，旨在为读者提供从理论到实践的全方位指导。书中涵盖了广泛的图像处理概念，包括但不限于图像采集、预处理、特征提取、图像分析、计算机视觉、数字信号处理以及机器学习在图像识别中的应用。作者们精心设计了章节结构，确保读者能够逐步掌握从基础原理到高级技术的过渡，无论是初学者还是专业研究人员都能从中受益。本书的特点在于其详实的理论讲解与丰富的实例演示，通过清晰易懂的语言，帮助读者理解复杂的算法和技术。同时，它也注重实际操作，提供了实用的代码示例和工具介绍，以便读者能够在实践中运用所学知识。版权方面，该书受到严格的保护，任何未经许可的复制或传播行为都必须得到Elsevier的书面授权。想要获取许可的读者可以通过拨打英国牛津的电话、发送传真或者电子邮件至Elsevier的科技版权部门进行申请，也可以通过Elsevier的官方网站在线完成请求流程。此外，该书已经向美国国会图书馆提交了分类记录，并且在大英图书馆也有相应的馆藏信息。ISBN号码为978-0-12-374457-9，表明这是一本经过专业编纂和校对的高质量著作。《图像处理基础指南2009ap》是一本不可或缺的图像处理学习和研究资源，对于想要深入了解并提升图像处理技能的读者来说，无论是在学术研究、工程实践还是教学工作中，都具有很高的参考价值。随着科技的发展，尽管部分内容可能已过时，但其基础知识和方法论仍然是宝贵的参考资料。

1.6 Quantized Images 11

a pixel

8-bit representation

FIGURE 1.9

Illustration of 8-bit representation of a quantized pixel.

individually or collectively (“vector quantization”);for example, a three-component color

image is frequently represented with 24 bits per pixel of color precision.

Unlike sampling, quantization is a difﬁcult topic to analyze since it is nonlinear.

Moreover, most theoretical treatments of signal processing assume that the signals under

study are not quantized, since it tends to greatly complicate the analysis. On the other

hand, quantization is an essential ingredient of any (lossy) signal compression algorithm,

where the goal can be thought of as ﬁnding an optimal quantization strategy that simul-

taneously minimizes the volume of data contained in the signal, while disturbing the

ﬁdelity of the signal as little as possible. With simple quantization, such as gray level

rounding, the main concern is that the pixel intensities or gray levels must be quantized

with sufﬁcient precision that excessive information is not lost. Unlike sampling, there is

no simple mathematical measurement of information loss from quantization. However,

while the effects of quantization are difﬁcult to express mathematically, the effects are

visually obvious.

Each of the images depicted in Figs. 1.4 and 1.8 is represented with 8 bits of gray

level resolution—meaning that bits less signiﬁcant than the 8

bit have been rounded or

truncated. This number of bits is quite common for two reasons: ﬁrst, using more bits

will generally not improve the visual appearance of the image—the adapted human eye

usually is unable to see improvements beyond 6 bits (although the total range that can

be seen under different conditions can exceed 10 bits)—hence using more bits would

be of no use. Secondly, each pixel is then conveniently represented by a by te. There are

exceptions: in certain scientiﬁc or medical applications, 12, 16, or even more bits may be

retained for more exhaustive examination by human or by machine.

Figures 1.10 and 1.11 depict two images at various levels of gray level resolution.

Reduced resolution (from 8 bits) was obtained by simply truncating the appropriate

number of less signiﬁcant bits from each pixel’s gray level. Figure 1.10 depicts the

256 ⫻ 256 digital image “ﬁngerprint” represented at 4, 2, and 1 bits of gray level resolu-

tion. At 4 bits, the ﬁngerprint is nearly indistinguishable from the 8-bit representation

of Fig 1.8. At 2 bits, the image has lost a sig niﬁcant amount of information, making the

print difﬁcult to read. At 1 bit, the binary image that results is likewise hard to read.

In practice, binarization of ﬁngerprints is often used to make the print more distinc-

tive. Using simple truncation-quantization, most of the print is lost since it was inked

insufﬁciently on the left, and excessively on the rig ht. Generally, bit truncation is a poor

method for creating a binary image from a gray level image. See Chapter 2 for better

methods of image binarization.

1.8 Size of Image Data 15

that are used in standard video camera systems: Red (R), Green (G), and Blue (B). The

way in which visible light wavelengths map to RGB camera color coordinates is a compli-

cated topic, although standard tables have been devised based on extensive experiments.

A number of other color coordinate systems are also used in image processing, printing,

and display systems, such as the YIQ (luminance, in-phase chromatic, quadratic chro-

matic) color coordinate system. Loosely speaking, the YIQ coordinate system attempts

to separate the perceived image brightness (luminance) from the chromatic components

of the image via an invertible linear transformation:

⎡

⎢

⎣

⎤

⎥

⎦

⫽

⎡

⎢

⎣

0.299 0.587 0.114

0.596 ⫺0.275 ⫺0.321

0.212 ⫺0.523 0.311

⎤

⎥

⎦

⎡

⎢

⎣

⎤

⎥

⎦

. (1.1)

The RGB system is used by color cameras and video display systems, while the YIQ is the

standard color representation used in broadcast television. Both representations are used

in practical image and video processing systems along with several other representations.

Most of the theory and algorithms for digital image and video processing has

been developed for single-valued, monochromatic (gray level), or intensity-only images,

whereas color images are vector-valued signals. Indeed, many of the approaches described

in this Guide are developed for single-valued images. However, these techniques are often

applied (sub-optimally) to color image data by regarding each color component as a sep-

arate image to be processed and recombining the results afterwards. As seen in Fig. 1.13,

the R, G, and B components contain a considerable amount of overlapping information.

Each of them is a valid image in the same sense as the image seen through colored spec-

tacles and can be processed as such. Conversely, however, if the color components are

collectively available, then vector image processing algorithms can often be designed that

achieve optimal results by taking this information into account. For example, a vector-

based image enhancement algorithm applied to the “cherries” image in Fig. 1.13 mig ht

adapt by giving less importance to enhancing the Blue component, since the image signal

is weaker in that band.

Chrominance is usually associated with slower amplitude variations than is lumi-

nance, since it usually is associated with fewer image details or rapid changes in value.

The human eye has a greater spatial bandwidth allocated for luminance perception

than for chromatic perception. This is exploited by compression algorithms that use

alternative color representations, such as YIQ, and store, transmit, or process the chro-

matic components using a lower bandwidth (fewer bits) than the luminance component.

Image and video compression algorithms achieve increased efﬁciencies through this

strategy.

1.8 SIZE OF IMAGE DATA

The amount of data in visual signals is usually quite large and increases geometrically

with the dimensionality of the data. This impacts nearly every aspect of image and

剩余840页未读，继续阅读

lamvda

粉丝: 0
资源: 5

2009年全面图像处理指南：学术出版社经典

The EssentialGuide to Image Processing

the essential guide to video processing

The Essential Guide to HTML5

The Essential Guide To RF And Wireless

The Essential Guide to HTML5.pdf

the essential guide to flex 2 3合集

The Essential Guide to Open Source Flash Development

The Essential Guide to Flash Games源代码

The Essential Guide to Flex 3 (Part 1)

The Essential Guide to 3D in Flash PDF+源码

最新资源