灰度图像彩色化的设计与实现毕业论文
2021-12-19 22:17:24
论文总字数:22746字
摘 要
灰度图像彩色化技术是利用计算机技术,将灰度图像或视频经过彩色化的处理,得到彩色图像和视频的过程。该技术的主要应用在于对一些历史影像进行修复和复原,和对军事、医学、航空等科学领域中非可见光成像的灰度图像进行特殊处理。因此,本次毕设的题目是灰度图像彩色化的设计与实现。
灰度图像彩色化的设计与实现采用深度学习技术从大量灰度图像和彩色图像的映射样本中学习彩色化知识,包括数据集建立、模型设计和训练、灰度图像彩色化测试等模块。数据集建立主要搜集灰度图像及对应的彩色图像,搜集的数据集尽可能涵盖光照、材质多种情况的变化。模型设计和训练主要运行基于本文的模型。灰度图像彩色化测试输入多张覆盖材质、光照、景物等关系变化的灰度图像,由调整、配置好的网络进行彩色化,验证彩色化结果。
该模型采用了VGG16的模型基础,采用迁移学习,结合了从已经训练的CNN神经网络和从Inception-ResNet-v2预训练模型中提取的高级特征来完善模型。同时本文,除了展示训练结果,接着还通过用户研究评估生成的图像的“公众接受度”。最后,提出了一个应用在不同类型和历史时期的图像,如历史照片的还原等等。
关键词:深度学习 灰度图像彩色化 CNN Inception-ResNet-v2 VGG16
迁移学习
Design and implementation of grayscale image colorization
Abstract
Grayscale image colorization technology is the process of obtaining color image and video by colorization of grayscale image or video using computer technology. The main application of this technology lies in the restoration and restoration of some historical images, and special processing of gray-scale images of non visible light imaging in military, medical, aviation and other scientific fields. Therefore, the topic of this design is the design and implementation of grayscale image colorization.
The design and implementation of grayscale image colorization adopts deep learning technology to learn colorization knowledge from a large number of mapping samples of grayscale image and color image, including data set building, model design and training, grayscale image colorization test and other modules. The establishment of data set mainly collects gray image and corresponding color image, and the collected data set covers as much as possible the changes of light and material conditions. Model design and training are mainly based on the model of this paper. The grayscale image colorization test inputs a number of grayscale images with different relations such as covering materials, lighting, scenery, etc., which are colorized by the adjusted and configured network to verify the colorization results.
The model is based on vgg16 model, and adopts migration learning. It combines the advanced features extracted from CNN neural network and Inception-ResNet-v2 pre- training model to improve the model. At the same time, in addition to displaying the training results, we also evaluate the "public acceptance" of the generated images through user research. Finally, we propose an image applied in different types and periods, such as the restoration of historical photos and so on.
Keywords: deep learning; grayscale image colorization; CNN; Inception-ResNet-v2; VGG16; migration learning
目录
摘要 I
Abstract II
第一章 绪论 1
1.1研究背景 1
1.2历史和起源 2
1.3国内外研究现状 3
1.4本文的结构安排 6
第二章 开发环境的搭建与介绍 7
2.1开发工具PyCharm 7
2.2 Pillow包简介 9
2.3 TensorFlow和Keras简介 9
2.4 Numpy包简介 10
第三章 模型及原理 14
3.1模型 15
3.2训练集 19
第四章 彩色化效果评估 21
4.1彩色图像经过灰度化后重新彩色化的评估 21
4.2用户评估 23
4.3老旧照片的还原效果 27
第五章 总结与展望 31
5.1 总结 31
5.2展望 32
参考文献 33
致谢信 36
第一章 绪论
1.1研究背景
从古代到今天,人们迎来了由黑暗到彩色的未来,图像也迎来了由黑白到彩色过程。颜色,作为图像信息的重要载体,相应的由此而衍生的灰度图像彩色化技术也是越发富有生机。现如今,图像处理技术广泛地应用于诸多领域,并且发挥着越来越重要的作用。同时,灰度图像彩色化技术也在发展。
灰度图像彩色化,通俗的讲就是为灰度图像或黑白图像这类缺乏颜色信息的图像上色。彩色化在航空航天,军事雷达,医学研究,卫星遥感等等方向有着很重要的研究价值。显而易见,在信息的丰富程度以及图像信息的交流上来说,彩色图像都要比灰度图像有着更好的表现效果。由此可见,彩色化是一项很有研究价值和意义的研究方向,如果能采用好的彩色化算法,不仅在视觉方面有着很大改善,对于信息的传递和还原也有着不可忽视的作用,并能根据所需来表达图像的原有意义。
请支付后下载全文,论文总字数:22746字