交互式现场鞋印花纹提取算法研究

发布时间：2018-06-08 13:41

本文选题：鞋印花纹提取 + 交互式图像分割　；参考：《大连海事大学》2017年硕士论文

【摘要】：鞋印是犯罪现场中遗留率较高的重要物证之一,在环境复杂的犯罪现场中准确提取鞋印花纹是足迹识别技术中急需解决的问题之一。鞋印花纹提取的关键工作是将花纹区域从复杂的背景中分割出来,与其他类型图像不同,现场鞋印图像种类多并且背景复杂多样,利用传统的图像分割方法很难得到理想的鞋印花纹分割结果。本文主要目的是对犯罪现场采集的鞋印图像进行处理,得到准确无噪声的鞋印花纹图像。交互式图像分割通过人机交互,在待分割图像上进行一定的目标和背景标记,对图像进行一定的先验信息的分割指导,然后根据先验信息建立分割算法模型,最后根据构建的模型得到目标分割结果。基于此目的本文对交互式现场鞋印花纹提取算法进行了研究,主要工作如下:1)给出了交互式现场鞋印花纹提取算法总体思路本文针对目前鞋印花纹提取算法存在的缺陷,给出了交互式现场鞋印花纹提取算法的总框架。该框架主要包括:结合鞋印图像方向特征的区域合并算法、基于kNN的交互式图像分割修正优化以及基于多尺度形态学的图像增强算法三大部分。2)给出了兼顾方向特性的区域合并算法根据分析现场鞋印图像的特性,本算法以传统的区域合并算法为基础,根据方向场特征和颜色特征相结合的相似性度量方法得到相邻区域的相似性,并结合用户的交互获取鞋印花纹的语义信息,给出了符合现场鞋印图像的区域合并方法。3)给出了基于kNN的交互式图像分割修正优化算法本文根据传统交互式图像分割的局限性,将鞋印花纹提取的过程逐步进行。对区域合并得到的花纹提取结果不满意时,利用kNN的分类策略,对花纹提取结果不断进行修正优化,直至得到较为满意的结果为止。经实验验证表明,这种修正优化算法减少背景干扰的同时,保留了鞋印花纹细节信息,提高了花纹的提取精度。4)给出了基于多尺度形态学的图像增强算法为得到准确清晰的鞋印花纹,本文给出了基于多尺度形态学的图像增强算法,该算法克服传统多尺度Top-hat算法的局限性,实现了不受光照影响的图像增强算法,最后利用二值化处理得到无背景干扰的清晰鞋印花纹图像。为验证交互式现场鞋印花纹提取算法的性能,本文运用包含151幅自然图像的数据集和从1189幅现场鞋印图像数据集随机抽取的100幅图像作为测试数据。本文运用4个评价指标判定分割准确性,经实验验证本算法的准确性达到了较高水平,是一种有效地现场鞋印花纹提取算法。
[Abstract]:Shoe printing is one of the important material evidence with high rate in the crime scene. It is one of the problems that need to be solved in the footprint recognition technology to extract the shoe print pattern accurately in the complex environment crime scene. The key work of shoe print pattern extraction is to segment the pattern area from the complex background. Different from other types of image, there are many kinds of shoeprint images on the spot and the background is complex and diverse. It is difficult to obtain ideal shoe print pattern segmentation results by using traditional image segmentation methods. The main purpose of this paper is to process the shoeprint image collected at the crime scene and get the accurate and noiseless shoeprint image. Interactive image segmentation is based on human-computer interaction, which carries out a certain target and background mark on the image to be segmented, and then gives guidance to the segmentation of the image with a certain prior information, and then establishes a segmentation algorithm model based on the prior information. Finally, the target segmentation results are obtained according to the constructed model. Based on this purpose, this paper studies the interactive spot shoe print pattern extraction algorithm. The main work is as follows: 1) the general idea of the interactive spot shoe print pattern extraction algorithm is given. This paper aims at the shortcomings of the current shoe print pattern extraction algorithm. The general frame of the interactive shoe print pattern extraction algorithm is presented. This framework mainly includes: combining the direction feature of shoeprint image, the region merging algorithm, The interactive image segmentation optimization based on KNN and the image enhancement algorithm based on multi-scale morphology. Based on the traditional region merging algorithm, the similarity of adjacent regions is obtained by combining the direction field feature and color feature, and the semantic information of shoe print pattern is obtained by user interaction. In this paper, an interactive image segmentation optimization algorithm based on KNN is presented. According to the limitation of traditional interactive image segmentation, the process of shoe print pattern extraction is carried out step by step. When the result of pattern extraction from region merging is not satisfied, the classification strategy of KNN is used to modify and optimize the result of pattern extraction until satisfactory results are obtained. The experimental results show that the modified algorithm can reduce background interference and retain the detail information of shoe print pattern at the same time. The image enhancement algorithm based on multi-scale morphology is given. In order to get accurate and clear shoe print pattern, the image enhancement algorithm based on multi-scale morphology is presented in this paper. The algorithm overcomes the limitations of the traditional multi-scale Top-hat algorithm and realizes the image enhancement algorithm which is not affected by illumination. Finally, the binarization processing is used to obtain the clear shoeprint image without background interference. In order to verify the performance of the interactive in-situ shoe print pattern extraction algorithm, this paper uses a data set containing 151 natural images and 100 randomly extracted images from 1189 on-site shoeprint image datasets as test data. In this paper, four evaluation indexes are used to judge the segmentation accuracy, and the accuracy of the algorithm is verified by experiments. It is an effective algorithm for shoe print pattern extraction in the field.
【学位授予单位】：大连海事大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：D918.91;TP391.41

【相似文献】