最新科研之《计算机视觉的展望及其观点》（小朱翻译）

前面的几章描述了一些关于计算机视觉以下的观点：

在关系数据库管理系统中，全局与局部的特性怎么产生，

以及如何快速索引到他们的以内容为主的索引算法怎么写

下面是这本书的内容主要概况：

第一章是整本书的一个介绍。

第二章讲的是几种检查和描述图像特性的方法，主要是从图像特征点出发，从边缘通过斑点的形式一点一点的展开到表现出它的全局特性。

第三章讲的是通过特征点的比较，来实现快速的对图片进行分类和检索

第四章讲的是一个新的特征描述的方法

第五章讲的是一系列的关系数据库实例

整体的总结及相应的结论：

    计算机视觉并不是一门成熟的学科，它仍然在发展及演变。因此，不可能在一本书里覆盖所有的方面和解决所有的问题。毫无疑问，视觉是我们最强大的感觉，所以在一般意义上很难有东西可以与之匹敌。然而，深度学习和硬件的飞速发展慢慢的改变了这个状况。在2015年，神经网络在大型图片识别比赛中打败了人类，计算机视觉开始从依赖人工标点到“自己”学习标点。用在第三章和第五章的训练特征的方法在未来是一个不错的研究方向，还有可能提高精准度呢。此外，在抗噪声、遮挡、失真、阴影等方面也可以得到不错的改善。计算机视觉的发展主要是因为计算机硬件的发展，因为许多NP算法是完备的。由于摩尔定理很有可能仍然有效，计算机视觉会变得越来越精确。

英文原文

The previous chapters covered some topics relating to computer vision: how global and local features are generated, how to fast index  them  and  how  to  imple- ment content-based retrieval algorithms in relational database management systems. Chapter 1 is an introduction to the book subject. Chapter 2 presents several meth- ods for image feature detection and description, starting from image interest points, through edge and blob detection, image segmentation till global features. Chapter 3 concerns feature comparison and indexing for efficient image retrieval and classifi- cation. Chapter 4 presents novel methods for feature description and Chap. 5 consists of a set of relational database implementation. Computer vision is not a mature disci- pline and is continually developing and evolving. Therefore, it is not possible to cover all the directions and solve all challenges within the scope of one book. Currently, it is hard to rival human vision in a general sense as it is our most powerful sense. Deep learning and hardware rapid development gradually change this situation. In 2015 neural networks defeated humans in the ImageNet Large Scale Visual Recognition Challenge. Computer vision starts to shift from relying on hand-made features to learned features. This can constitute a direction in the future research, namely, using trained features in the methods described in Chaps. 3 and 5, would possibly improve the accuracy. Moreover, the robustness in terms of immunity to noise, occlusions, distortion, shadows etc. can also be improved. Computer vision benefits heavily from the development of computer hardware as many algorithms are NP-complete. Since Moore’s law (and other types of the hardware development) will most likely still be valid, vision system will be more and more sophisticated.