英文原文
2011 Eighth International Conference on Fuzzy Systems and Knowledge Discovery (FSKD)
An Approach of Color Feature Evaluation in Color
Recognition
Qi Xiaoxuan, Ji Jianwei
School of Information amp; Electric Engineerin
Shenyang Agriculture University
Shenyang, China
Abstract—This paper analyzes the characteristics of five commonly used color spaces and explores their influences on color recognition respectively. Divisibility evaluation based on distance criterion is utilized to evaluate the different colorfeatures in each color space and experimental results show that HSI color space has the best divisibility performance. Keywords-color space;colorrecognition; feature evalutation; divisibility critiron
I. I NTRODUCTION
Color is the most intuitive vision feature to describe colorful images. It has been widely used in pattern recognition for the reason that color feature is almost free from the effects of scale, rotation and translation for the input images [1]. Colors in colorful images can be defined by different color space models, such as RGB space, CMY space, I1I2I3space, YUV space and HSI space. Among the above color spaces, RGB is the basic and the most common one and can readily be mapped into other color spaces. However, RGB space is non-uniform for
color perception and is too easily influenced by light. The three color components of RGB space are correlated with each other [2]. CMY space represents colors by the complementary component of RGB components. YUV space, frequently used in color TV systems, uses three channels as Y, U and V to define the pixel. Y are the brightness information, U and V are the color difference which denotes the overall color difference instead of the difference between the three components of RGB. HSI space is a uniform one which consists to the human perception of colors. Its three components are mutually independent and can perceive color change of each component respectively. But non-linear transform in HSI space may lead substantial computation as well as singularity of the color space when the saturation is low. While in YCbCr color space, the chrominance component and the luminance component are interdependent. Besides that, the conversion from YCbCr space to RGB space is linear and simple, so YCbCr space is commonly used in the field of video encoding compression. YUV space, YCbCr space and HSI space all represent spectrum in two dimension and use the third dimension to represent the intensity
of color, which enables them more suitable for occasions where light intensity changes, than RGB space.Color recognition technique has been applied to many fields and has gone ahead rapidly. For instance, color recognition in product surface, license plates identification, face recognition and skin recognition [3-6]. Color recognition effects differ with the change of color space. This paper investigates on color feature divisibility in the commonly used color spaces as RGB space, CMY space, YUV space, YCbCr space, I1I2I3space and HSI space. Analysis indicates that HSI has the best divisibility performance in all the
above color spaces based on the distance criterion. It provides a theory basis for color recognition.
II. COLOR SPACE AND I TS T RANSFORMATION
It is essential to build up and select a suitable color space for obtaining a kind of valid color features to characterize colorful images. Different color spaces are utilized for different research purposes. Color space means to define color by an
array in three-dimension space. In the processing of colorful images, color space is also named as color model or color coordinates. One color space can be converted to another by certain transforms. Below is the introduction of some color spaces and their conversions [7].
A. RGB Color Space
Red (R), green (G), blue (B) are three primary colors ofspectrum. All colors can be generated by the sum of the threeprimary colors. In digital images, values of R, G and B rangefrom 0 to 255. A cube in three-dimension coordinate space can be used to describe the RGB color space, where red, green andblue are the three axes, shown in Fig.
1.The main drawback of RGB color space as follows:
bull; It is not intuitive. It is difficult to see from the RGBvalues the cognitive attributes that the color representsitself.
bull; It is non-uniform. The perception difference betweentwo colors in RGB space is different from the distancebetween the two colors.
bull; It is dependent on hardware devices.
In a word, RGB space is device-related and an incompleteintuitive color description. To overcome these problems, othercolor spaces,which are more in line with characteristics of color vision, are adopted. RGB space can be mapped to other color spaces readily.
B. CMY(CMYK) Color Space
CMY space is a spatial structure of a rectangular Cartesian. Its three primary components are cyan (C), magenta (M) and yellow (Y). Colors are obtained by subtractive colors. CMY space is widely used in non-emission display as inkjet printers. Equal amount of the three components can generate the black color. But the aforementioned black color is not pure. Generally speaking, to generate true black color, the fourth component, i.e. black, is added in. This is the CMYK color space. CMY space is not very intuitive and non-linear. Its three components are the complementary colors of R, G and B. The transformations are as follows:
The transformations from RGB space to CMY space are as follows:
C. YUV and YCrCb Color Space
YUV space and YCbCr space both generate a luminance component and two
chrominance c
剩余内容已隐藏,支付完成后下载完整资料
关于颜色识别中颜色特征分析的方法
摘要:分析五种常用的颜色空间的特征并研究其分别对颜色识别的影响。基于距离标准的可除性鉴定被用来评定每个颜色空间内的颜色特征,实验结果表明HSI颜色空间可除性最强。
关键词:颜色空间;颜色识别;特征分析;可除性标准
1. 概述
颜色是描述色彩图像时最直观的视觉特征。鉴于颜色特征几乎不受范围,旋转及转化对输入图像的干扰,颜色被广泛地应用于图像识别[1]。色彩图像中的颜色可由不同的颜色空间模式规定,比如RGB空间,CMY空间,I1I2I3空间,YUV空间和HSI空间,其中,RGB是最基本也是最常见的颜色空间,并且可以很容易的映入其他颜色空间中。但是,RGB空间与颜色直觉不一致,而且过于容易被光线影响。此空间的三个颜色分量是互相关联的[2]。CMY空间通过RGB空间分量补充性分量来表现颜色。常被用于彩色电视系统的YUV空间通过Y,U,V三个波道来定义像素。Y表示的是亮度信息,U和V是色差,并决定整体色差,而RGB则是靠三个分量之间的差别来影响整体色差的。HIS空间与人类对颜色的直觉相一致,其三个分量互相独立,并且可以分别觉察到每个分量的变化,但是HSI空间中的非线性变化可能导致当饱和度低的时候出现大量的运算及颜色空间的异常。然而在YCbCr颜色空间中,色度成分和亮度成分是相互依赖的。除此之外,由于从YCbCr空间到RGB空间的转换是很简单的线性转化,所以前者通常被应用于视频编码压缩领域。YUV空间,YCbCr空间和HSI空间都用二维来表现光谱,用第三维来表现颜色的强度,这使他们比RGB空间更容易适应光线强度的变化。颜色识别技术现已被应用于各项领域,发展迅速;如产品表面,汽车牌照定位,人脸及皮肤纹理识别[3-6]。随着颜色空间的变化,颜色识别的影响也会有所不同。本文通过研究RGB空间,CMY空间,YUV空间,YCbCr空间,I1I2I3空间及HSI空间这几个常用颜色空间的颜色特征可除性,得出结论:在上述基于距离判据的颜色空间中,HSI最具可除性;并为颜色识别提供理论基础。
2. 颜色空间及其转化
建立并选择一个适合的颜色空间对获得一种有效的颜色特征来描述色彩图像的特征至关重要。不同的颜色空间用于不同的研究目的。颜色空间是指用三维空间的数组来定义颜色。在色彩图像的处理过程中,颜色空间又称颜色模型,或颜色坐标。通过某些转化,可以将一个色彩空间转换为另一个。下面是一些颜色空间的介绍及其转化[7]。
2.1 RGB颜色空间
光谱的三个原色是红(R),绿(G),蓝(B),所有的颜色都能用三种基本组合起来形成。在数码影像中,R,G,B的取值范围0到255。在三维坐标空间中的多维数据集可用于描述 RGB 颜色空间,红,绿,蓝分别为轴线,如图表1所示。
RGB颜色空间最主要的缺点如下所列:
1. 非直观性。很难从RGB数据库中看到颜色本身所表示的认知属性。
2. 非一致性。RGB颜色空间中两种颜色间的感知区别与其间的距离是不同的。
3. 依赖于硬件设备
总之,RGB空间与设备相关,是一个不完备的直观颜色说明。为了解决这些问题,采用其他更符合颜色视觉特征的颜色空间,使得RGB空间可以很容易映射到其中。
2.2 CMY(CMYK)颜色空间
CMY 空间是笛卡尔直角坐标系的空间结构,其三个主要组件是青色(C),品红(M),黄色(Y)。相减混色模式以吸收三基色比例不同而形成不同的颜色。CMY空间被广泛应用于像喷墨打印机一类的非发射显示器。理论上来说,三种等量基色可形成黑色,但是由于目前制造工艺还不能造出高纯度的油墨,CMY相加的结果实际是一种暗红色。一般而言,要想得到高纯度的黑色就要加入第四个基色,也就是黑色;这就是CMYK颜色空间。CMY 空间不是很直观且非线性,它的三个组件是红,绿,蓝三个互相补充的颜色,在CMY模型中,颜色是从白光中减去一定成分得到的。其转化如下所示:
(1)
从RGB空间到CMY空间的转化如下所示:
(2)
2.3 YUV和YCbCr颜色空间
YUV空间和YCbCr空间都含有一个亮度元素和两个色度元素。在YUV空间中,Y表示明亮度,而U和V表示的则是色差,它的亮度信号Y和色度信号U、V是分离的,而且,YUV 空间可以减少由人类视觉特性的数字彩色图像所需的存储容量。在YCbCr空间中,Y是指亮度分量,Cb指蓝色色度分量,而Cr指红色色度分量;其优点显而易见:颜色分量与亮度元素相分离,并可从RGB空间进行线性转换。通过以下方程式可大致解释从RGB到YUV的转换关系:
(3)
2.4 HSI颜色空间
它反映了人的视觉系统感知彩色的方式,以色调(H)、饱和度(S)和强度(I)三种基本特征量来感知颜色;与其相似的还有HSV颜色空间(色相h,饱和度s,色调v)和HSB颜色空间(色相h,饱和度s,亮度b),这些都属于极坐标空间的结构,它们共同的优点是可以直观的描述颜色,且大多可以从RGB空间进行线性转换。HSI模型的建立基于两个重要的事实: 一是I分量与图像的彩色信息无关;二是H和S分量与人感受颜色的方式是紧密相联的,其中H分量对颜色的描述能力最接近于人类的视觉,因此,它的区分力也是最强的[8]。 RGB空间到HSI空间的转换如下方程式所示:
(4-6)
有三元素的HSI颜色空间与人类习惯相适应,可以更好的描述颜色。但色差中依然存在非线性的缺点,尤其是H分量的颜色与角度[9]。
2.5 I1I2I3颜色空间
RGB 空间到I1I2I3 空间的线性转换如下方程式以得到三个颜色直方特征:
(7)
由公式(7)可知,I1,I2,I3 的数值可正可负,在图像识别中,I1I2I3空间具有最佳的非相关性。
3. 颜色空间的特征分析
通过颜色空间,抽象、主观的视觉感知可被转化为三维空间中的特定位置、矢量,从而有可能使彩色图像和设备的颜色特征可视化。颜色空间是颜色识别的重要工具,各种混合系统有其相应的颜色空间并有不同的属性和各自的优缺点。颜色空间的有效性是处理彩色图像的关键,可除性标准则被用来检测不同颜色空间对颜色的分类。距离判据具有简明清晰的概念,即,类属中距离越小,类属间距离越大,可除性便越大。下面的特征分析运算法则便是基于距离判据[10]。 计算均值向量和i 类样本的协方差,N 是总样本数,Ni 是i 类样本数。
(8-9)
计 算 总 的 均 值 向 量和协方差 。
(10-11)
构造i 类 样 本 的 散 射 矩阵。
(12)
构 建 在 同一类 样 本 的 总 散 射 矩阵。
(13)
构 建 不同的 类之间 样 本 的 总 散 射 矩阵。
(14)
定义评价指标,如下所示:
(15)
4. 实验结果及分析
颜色被人的眼睛分为十一类,红色、绿色、蓝色、黄色、紫色、橙色、粉红色、棕色、灰色、白色和黑色,如图2。
对RGB空间,CMY空间,YUV空间,I1I2I3空间,和HSI空间分别进行评估运算,特征参数和评价指标如表格I所示。
从表格I 可知,HSI 空间与其它四个所分析的颜色空间相比最具可除性。
5. 总结
必须选择有效的颜色空间来处理彩色图像。本文在可除性标准的基础上,对五个常用的颜色空间进行分析和比较,实验结果表明HSI 颜色空间具有最强的可除性。这也为颜色识别中的颜色空间选择提供了基础。
致谢
本项目由辽宁省自然科学基金资助(项目编号:20102153)。
剩余内容已隐藏,支付完成后下载完整资料
资料编号:[23109],资料为PDF文档或Word文档,PDF文档可免费转换为Word
课题毕业论文、外文翻译、任务书、文献综述、开题报告、程序设计、图纸设计等资料可联系客服协助查找。