基于图像处理的服装模特图片检测分割与识别研究

发布时间：2018-06-22 16:50

本文选题：图像分类 + 图像识别　；参考：《深圳大学》2016年硕士论文

【摘要】：近些年随着互联网的迅速发展与普及,网购已成为人们生活的一部分,服装更成为人们网购中的热点。图像处理技术在电子商务中的应用成为了一个研究的热点,比如在服装销售领域,模特图片如何按照网购网站的展示规则自动裁剪,如何自动识别图片中的服装生成相应的标签标示衣服的类型,以及如何把模特从背景图像中分离出来等等。本文就从这几个方面展开研究,内容如下:1.图像预处理。通过分数阶微分算子进行图像增强、中值滤波等一系列图像预处理方法把服装模特图片边缘加强,用Canny算子提取出服装模特的边缘信息。利用人脸检测定位出人脸位置,提取出含有模特的核心区域。2.图像分类。通过不同服装模特图片中衣服类型的不同,其相应的特点也不同,提取出最有利于区分服装类别的特征,本文提取了图像二值化后特定段边缘点的数目特征和分布特征、图像角点的分布特征以及图像边缘的方向特征,再通过最小距离分类器的方法进行分类。3.图像识别。通过服装模特图片中衣服在YCr Cb颜色空间下的直方图信息提取出颜色特征、通过图像灰度共生矩阵信息提取出纹理特征、通过图像角点检测提取出形状特征等识别出不同类型衣服所具有的不同特点,从而自动生成相应的标签。4.图像分割。通过确定自适应阈值进行二值化处理,再通过Canny算子进行边缘检测,把二者相结合,并通过多次膨胀和腐蚀改善噪声,精确目标区域的边缘,把模特从背景中分离出来,并可选择其他背景进行背景替换。在研究的过程中本文共提出以下创新点和改进:1.对于物体的边缘识别提出先分类后识别的方法。2.对于网购服装提出根据图像识别,提取出相应特征,自动生成标签。3.对于图像分割提出自适应阈值分割和边缘检测相结合的方法。本文把服装模特图片的按规则裁剪问题转化成先按服装类型进行分类,再根据相应类型特点找到服装下摆位置的问题,从而完成裁剪。本文根据图片中衣服颜色特征、纹理特征、形状特征等识别出图片中的服装的相应特征并自动生成标签保存。本文还完成了把人物从背景图像中分离出来,并重新选择新的背景进行替换。
[Abstract]:In recent years, with the rapid development and popularity of the Internet, online shopping has become a part of people's lives, clothing has become a hot spot in online shopping. The application of image processing technology in electronic commerce has become a research hotspot, for example, in the field of clothing sales, how model pictures are automatically cut according to the display rules of online shopping website. How to automatically identify the clothing in the image and how to separate the model from the background image and so on. This article carries on the research from these several aspects, the content is as follows: 1. Image preprocessing. A series of image preprocessing methods, such as fractional differential operator and median filter, are used to enhance the image edge of clothing model, and Canny operator is used to extract the edge information of clothing model. Using face detection to locate the position of the face, extract the core region containing models. 2. Image classification. According to the different clothing types of different clothing models, their corresponding characteristics are also different, which is most helpful to distinguish the clothing categories. In this paper, the number and distribution characteristics of the edge points of a particular segment after image binarization are extracted. The distribution feature of image corner and the direction feature of image edge are classified by the method of minimum distance classifier. Image recognition. The color feature is extracted from the histogram information of clothes in YCr CB color space, and the texture feature is extracted from the image gray level co-occurrence matrix information. The shape features of different types of clothes are extracted by image corner detection and the different characteristics of different types of clothes are identified, and the corresponding labels. 4. 4 are generated automatically. Image segmentation. By determining the adaptive threshold for binary processing, and then using Canny operator for edge detection, combining the two, and through multiple expansion and corrosion to improve the noise, accurate the edge of the target area, the model is separated from the background. Other backgrounds can be selected for background replacement. In the course of the research, this paper puts forward the following innovations and improvements: 1. For edge recognition of objects, a method of classification and recognition is proposed. For online clothing, according to the image recognition, extract the corresponding features, automatically generate label. 3. An adaptive threshold segmentation and edge detection method is proposed for image segmentation. In this paper, the problem of pattern cutting of clothing models is transformed into classification according to clothing types first, and then the problem of dress pendulum position is found according to the characteristics of the corresponding types, so as to complete the clipping. According to the color features, texture features and shape features of the clothes in the picture, this paper identifies the corresponding features of the clothing in the picture and automatically generates labels to save them. This paper also completes the separation of the characters from the background image and the re-selection of the new background to replace it.
【学位授予单位】：深圳大学
【学位级别】：硕士
【学位授予年份】：2016
【分类号】：TP391.41

【相似文献】