光照变化下鲁棒的颜色分割算法
176 浏览量
更新于2024-08-30
收藏 773KB PDF 举报
"鲁棒光照变化条件下的色彩分割算法"
这篇论文介绍了在光照变化条件下的一种新型的、对光照变化具有鲁棒性的色彩图像分割方法，特别应用于皮肤分割领域。该方法基于色调保护算法，旨在解决光照变化对图像分割效果的影响，从而提高智能识别率。
在传统的三原色（RGB）空间中，光照变化会导致像素颜色的变化，进而影响图像分割的准确性。为了解决这个问题，该方法将RGB空间降维到一维，同时保持每个像素在空间变换前后不变的色调。通过这种降维处理，可以在新的颜色空间中建立皮肤颜色模型，通常采用高斯模型来表示。降维操作不仅减少了计算复杂性，还能确保颜色信息的关键特征——色调——得以保留。
实验结果显示，这种方法在光照变化下具有很好的鲁棒性，能有效抵抗光照条件变化对图像分割结果的影响。同时，由于其较低的计算复杂性，该算法在实际应用中也具有较高的效率和实用性。
该研究的相关光学信息代码为100.2000和100.3008，表明它属于光学和图像处理的范畴。论文的doi为10.3788/COL20100803.0286，可以用于进一步查找和引用该研究成果。
这篇论文的核心知识点包括：
1. **光照变化对图像分割的影响**：光照变化会导致图像颜色的变化，影响图像分割的准确性和智能识别的性能。
2. **色调保护算法**：通过保持像素的色调不变，即使在空间转换后，也能保持颜色信息的完整性。
3. **RGB空间降维**：将三维的RGB空间压缩为一维，减少计算复杂性，同时保持关键的色调信息。
4. **皮肤颜色模型**：在新颜色空间中，使用高斯模型建立皮肤颜色模型，以进行有效的皮肤区域分割。
5. **鲁棒性与计算复杂性**：所提出的方法在光照变化条件下表现稳定，并且计算复杂度较低，适用于实时或资源受限的系统。
6. **光学信息代码**：100.2000和100.3008分别对应特定的光学研究领域，如光学成像和色彩科学。
7. **实验验证**：实验结果证明了该方法在不同光照条件下的有效性和可靠性。
这些知识点对于理解光照变化条件下图像处理的挑战，以及开发鲁棒的色彩分割算法至关重要，对于计算机视觉、图像处理和机器学习等领域具有重要的理论和实践价值。
286 CHINESE OPTICS LETTERS / Vol. 8, No. 3 / March 10, 2010
Robust color segmentation algorithms in illumination
variation conditions
Jinhui Lan (777777)
∗
and Kai Shen ( ppp)
Department of Measurement and Control Technologies, School of Information Engineering,
University of Science and Technology Beijing, Beijing 100083, China
∗
E-mail: jh.lan@263.net
Received June 15, 2009
Changing illumination condition can change the result of image segmentation algorithm and reduce the
intelligent recognition rate. A novel color image segmentation method robust to illumination variations
is presented. The method is applied to the skin segmentation. Based on the hue preserving algorithm,
the method reduces the dimensionality of the red-green-blue (RGB) space to one dimension, while keeping
the hue of every pixel unchanging before and after space transformation. In the new color space, the skin
color model is established using Gaussian model. Experimental results show that the method is robust to
illumination variations, and has low computational complexity.
OCIS co des: 100.2000, 100.3008.
doi: 10.3788/COL20100803.0286.
Color segmentation is an essential, critical, and prelimi-
nary process in a lot of vision-based tasks such as visual
tracking, human-computer interaction (HCI), human-
robot interaction (HRI), vision-based robotics, visual
surveillance, and so forth, because color is an effective
and robust visual cue for characterizing an object from
the others. Recently, there has been a growing interest
in the skin segmentation, which aims at detecting human
skin regions in an image.
However, changing illumination condition can change
the characteristics of a color, and limit the applications
of the color segmentation
[1]
. Therefore, a lot of re-
searches have been carried out for invariant detection of a
color under illumination-variation conditions. Static skin
model based methods were adopted
[2−5]
, which firstly
selected an appropriate color space. The color space
transformation has been assumed to increase separabil-
ity between skin and non-skin classes to increase sim-
ilarity among different skin tones. Then a model of
skin color in the selected color space is established to
decide whether a pixel belongs to a skin region. The
color space mainly includes normalized red-green (RG)
[2]
,
hue-saturation-value (HSV) (hue-saturation-illumination
(HSI))
[5]
, YCrCb
[6]
, YIQ
[7]
, YES
[8]
, etc. The skin color
model includes single Gaussian model
[2]
, Gaussian mix-
ture models
[3]
, Gaussian mixture models considering in-
tensity information
[5]
, etc. Cho et al. and Soriano
et al. used dynamic learning based methods for color
segmentation
[9,10]
. These methods dynamically allocate
a color through various illumination conditions while
those models are updated for every image sequence.
Adaptive skin color filter and ‘skin locus’ adaptive skin
color modeling are two typical methods of dynamic learn-
ing based methods.
In this letter, a novel color image segmentation method
which is robust to illumination variations is proposed.
Since RGB space is the most common space to represent
color images, and HSI space can separate the chromi-
nance and luminance in a color image to a certain extent,
we consider both the characteristics of RGB space and
HSI space, and reduce the dimensionality of the RGB
space to one dimension while preserving the hue of ev-
ery pixel in the image before and after space transforma-
tion constant by hue-preserving algorithm. Therefore,
1D threshold can be used to sub divide the color image.
Experimental result verifies that the proposed method is
not only robust to the changing illumination condition,
but also more efficient in computation efforts.
Hue preservation is necessary when converting colors
from one color space to another. Distortion may o ccur
if hue is not preserved. The hues of a pixel in the scene
before and after the transformation should be the same
for a hue preserving transformation. First of all, a gen-
eral hue-preserving method between RGB space and HSI
space is introduced for color image segmentation.
The conversion equations from the RGB space to HSI
space are noted as
H = arctan
Ã
√
3 (G − B)
(R − G) + (R − B)
!
S = 1 −
3 [min (R, G, B)]
(R + G + B)
I =
1
3
(R + G + B)
. (1)
In general, color images are stored and viewed using
RGB space. To pro cess an image for segmentation in
HSI space, the image needs to be transformed. As can
be seen from Eq. (1), this transformation is computation-
ally costly
[11]
, and the inverse coordinate transformation
has to be implemented for displaying the images.
Two hue-preserving operations, scaling and shifting,
were introduced in Ref. [12] for luminance and saturation
processing. Using these two operations, hue-preserving
method for color image segmentation is developed.
We denote the grey values for R, G, and B components
of an image pixel I by a vector x, where x = (x
1
, x
2
, x
3
),
x
1
, x
2
, and x
3
correspond to the red, green, and blue
pixel values, respectively. That is, 0 ≤ x
k
≤ 255, k = 1,
2, and 3.
Scaling the vector x to x
0
by a factor α > 0 is defined as
1671-7694/2010/030286-04
c
° 2010 Chinese Optics Letters
下载后可阅读完整内容，剩余3页未读，立即下载
2016-03-02 上传
2020-04-06 上传
2021-02-10 上传
2016-11-04 上传
2011-01-19 上传
2021-10-02 上传
2021-06-29 上传
2021-02-10 上传
2017-10-24 上传
weixin_38592420
- 粉丝: 6
- 资源: 935
我的内容管理
展开
最新资源
- 前端协作项目：发布猜图游戏功能与待修复事项
- Spring框架REST服务开发实践指南
- ALU课设实现基础与高级运算功能
- 深入了解STK：C++音频信号处理综合工具套件
- 华中科技大学电信学院软件无线电实验资料汇总
- CGSN数据解析与集成验证工具集：Python和Shell脚本
- Java实现的远程视频会议系统开发教程
- Change-OEM: 用Java修改Windows OEM信息与Logo
- cmnd：文本到远程API的桥接平台开发
- 解决BIOS刷写错误28：PRR.exe的应用与效果
- 深度学习对抗攻击库：adversarial_robustness_toolbox 1.10.0
- Win7系统CP2102驱动下载与安装指南
- 深入理解Java中的函数式编程技巧
- GY-906 MLX90614ESF传感器模块温度采集应用资料
- Adversarial Robustness Toolbox 1.15.1 工具包安装教程
- GNU Radio的供应商中立SDR开发包：gr-sdr介绍
