多角度多尺度鲸尾图像检索系统的研究与实现

发布时间：2018-04-21 01:09

本文选题：鲸尾图像 + 局部特征描述子　；参考：《西北农林科技大学》2017年硕士论文

【摘要】：鲸鱼作为海洋食物链顶端的一环,在整个生态系统中的地位举足轻重,近年来却处于濒危的境地。因此研究人员需要追踪监测鲸类的年龄、生长速度、繁殖周期及其种群状况对鲸类的分布规律、生态环境、种群特征以及引起濒危的原因等进行研究。自动化的鲸鱼个体识别对这项科研工作有着重大意义。本研究基于多角度与多尺度空间实现鲸尾图像的检索,首先构建鲸尾图像的多角度(多视角)空间,然后将具有尺度不变性的局部特征提取算法应用于该多角度图像空间,提取出具有旋转、平移以及仿射不变性的多角度多尺度鲸尾图像特征,。本文的主要研究内容如下:(1)介绍并分析鲸尾图像数据集。针对本研究数据集中的鲸尾图像由多视角采集的特点构造输入图像的多角度(多视角)空间以模拟实现不同角度拍摄目标图像所发生的畸变。(2)对多角度多尺度鲸尾图像特征提取和匹配的研究。基于多角度多尺度的思想并参照Affine-SIFT算法的实现原理,本文研究了Affine-SURF、Affine-BSIFT特征提取算法。使最终提取的鲸尾图像局部特征具有仿射、旋转和平移不变性,适用于多视角采集的鲸尾图像。(3)选取多角度多尺度鲸尾图像特征提取算法。实验分析了Affine-SIFT、Affine-SURF以及Affine-BSIFT算法应用于鲸尾图像的性能。进行比较之后,基于稳定性的需求选择Affine-SIFT算法提取鲸尾图像特征。将Affine-SIFT算法进行优化后应用于鲸尾图像数据集,当检索的相似鲸尾图像数据集的大小取10时,其成功率0.48。(4)设计并实现了多角度多尺度的鲸尾图像检索系统。在进行需求分析之后,基于多角度多尺度鲸尾图像特征提取算法和匹配方法初步设计和实现了多角度多尺度的鲸尾图像检索系统。
[Abstract]:As the top of the marine food chain, whales play an important role in the whole ecosystem, but are endangered in recent years. Therefore, researchers need to track and monitor the age, growth rate, breeding cycle and population status of cetaceans to study their distribution, ecological environment, population characteristics and causes of extinction. Automated identification of individual whales is of great significance to this scientific work. In this paper, the retrieval of whale tail image is realized based on multi-angle and multi-scale space. Firstly, the multi-angle (multi-angle) space of whale tail image is constructed, and then the scale-invariant local feature extraction algorithm is applied to the multi-angle image space. The features of multi-angle and multi-scale whale tail images with rotation, translation and affine invariance are extracted. The main contents of this paper are as follows: 1) introduce and analyze the whale tail image data set. According to the characteristics of the whale tail image in the data set of this study, the multi-angle (multi-angle) space of the input image is constructed to simulate the distortion of the target image taken at different angles. Research on feature extraction and matching of tail image. Based on the idea of multi-angle and multi-scale and referring to the realization principle of Affine-SIFT algorithm, the feature extraction algorithm of Affine-SURFU Affine-BSIFT is studied in this paper. The local features of the extracted whale tail image are affine, rotated and translational invariant, which is suitable for the multi-angle and multi-scale whale tail image feature extraction algorithm. The performance of Affine-SIFTF-Affine-SURF and Affine-BSIFT algorithm applied to whale tail image is analyzed experimentally. After comparison, the requirement selection Affine-SIFT algorithm based on stability is used to extract the features of whale tail image. The Affine-SIFT algorithm is optimized and applied to the whale tail image data set. When the similar whale tail image data set is retrieved at 10:00, its success rate is 0.48.44) and a multi-angle and multi-scale whale tail image retrieval system is designed and implemented. After requirement analysis, a multi-angle and multi-scale whale tail image retrieval system is designed and implemented based on multi-angle and multi-scale whale tail image feature extraction algorithm and matching method.
【学位授予单位】：西北农林科技大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP391.41

【参考文献】