Second week--team project (small big guy with flying team)

Name	School Number
Yang Hao (group leader)	3116004705
Shei Min	3116004478
Zhong Wei	3116004711
Li Yiju	3116004691
Li Xiaoyang	3116004641
Sho Jie (small)	3216004492

Text image recognition is one of the important research topics in machine vision field. Text image recognition covers many applications and tasks, most of which originate from the digitization of documents decades ago. It has become an extremely important research content in the field of information processing to realize its classification, recognition and retrieval by using RAW image data. The user is able to carry on the higher level operation and the understanding to the image document, therefore the text image recognition technology obtains the more and more widespread attention and the application.

Requirements Specification Description Basic project requirements

(1) Identification of different types of single-image inspection

Image preprocessing can improve the reliability of feature extraction, image segmentation, matching and recognition by eliminating extraneous information in the image, recovering useful real information, enhancing the detectable nature of information and minimizing the data. Therefore, the selection of pre-processing algorithms for the verification of single images is particularly critical. Then through the operation of the feature extraction of the single image, the neural network model is used to identify the training and improve the recognition rate.

(2) Classification of the ROI portion of the data availability area in the verification form

the ROI Region processing and analysis can not only reduce the complexity of the analysis process, but also reduce unnecessary computational waste, because each of the data area is fixed, so we decided to use the threshold segmentation algorithm in the global threshold method and canny operator and other edge detection algorithm to extract ROI region and image segmentation.

(3) feature extraction and recognition of data effective regions

The procedure of detecting the text information of valid area of data is usually to locate the image text, to separate the text from the background, to obtain the text information, to detect the data region by using the mser algorithm and the stroke width transform algorithm in the text detection. The paper uses convolutional neural network alexnet model to train and classify the sample characteristics in terms of word recognition with support vector machine.

(4) Flowchart

Advanced Project Requirements (1) Provide graphical interface to users

　　The basic requirement only needs to provide the interface to the user to get the detected text data from that interface, but when the type of verification is increased, the model needs to be updated manually to satisfy the verification of the order. Another more difficult solution is to provide users with a graphical interface, you can add a new type of picture on the graphical interface, input tags, and provide an interface to the original model based on the iteration of the new model, making the whole system more automated and convenient.

About the system

(1) Authenticity

The image data processed by this project is derived from the output inspection of the eyeball inspection equipment.

(2) Availability

This project provides a well-encapsulated interface for use by the user, enabling it to obtain a text format file for the validation of a valid data area after the input of the verification picture and continue the next operation of the data. By eliminating the tedious operation of manual entry of medical personnel, the doctor's efficiency is greatly improved.

(2) The value of

Eyeball, as an important part of human biological characteristics, contains abundant information, which has been paid close attention to by researchers at home and abroad in recent years. At the same time, eye recognition technology has also been rapidly developed. Eye recognition can accurately obtain eye information, such as the location of the eye, pupil center, Iris Information and even the direction of the line of sight. The use of image processing and pattern recognition technology to identify the eyeball inspection and data areas, you can further processing the information, extract all kinds of corresponding data, so that the diagnosis and treatment process more automated, saving patients and doctors time, improve efficiency.

Project Specifications

(1) Project adoption Code Cloud management platform: project address

(2) To develop the identification quantity: Because in the initial stage, the test data set is relatively small, tentatively 20 categories of verification.

Code specification

1. Module names should be compatible with their functions, each module should be annotated at the beginning of the file module functions and Precautions
2. Write docstrings for all common modules, functions, classes, methods; Not common, not necessary, but can write comments (on the next line of Def)
3. The module is named as short as possible, using the all lowercase method, you can use underscores.
4. The package name is as short as possible, using all lowercase methods, and the underscore cannot be used.
5. The class is named using Capwords, and the classes used inside the module take the _capwords approach.

Code Cloud Issues Page

Project Plan table

6th Week	1. Team teaming, Team blog (?)
	2. Team presentation, member presentation, role assignment, selection of selected topics (?)
	3. Develop team planning arrangements (?)
7th Week	1. Requirements Specification Sheet
	2. Prototype design, team members estimate task difficulty and learn the necessary techniques
	3. Coding specification Complete, platform environment set up, preliminary construction
8th Week	1. Prototype improvement (to demonstrate the prototype to the target user and further understand the requirements)
	2. Architecture design, WBS, team members estimate the time required for their respective tasks
	3. Test Plan
9th Week	1. Team Project Alpha Task assignment schedule
	2.7 consecutive days of Alpha Agility Sprint, 7 daily Scrum meeting blog + code submission
10th Week	1. User feedback + test plan improvement
	2. Team Alpha Stage Personal summary
	3. Team Project Alpha Blog: Release notes, test reports, presentation blogs, project management
11th Week	1. Team Project Alpha Blog: Post-mortem analysis
	2. One person on each team must leave and look for the next team to accept. The team sent out blogs announcing the departure and acceptance of the members.
12th Week	1. Team Project Beta Task Assignment plan, introducing new members
	2.7-day Beta Agile Sprint, 7 daily Scrum meeting blog + code submission
13th Week	1. Team Project Beta Blog: Release notes, test reports, show blogs
	2. Team Beta Stage Personal summary
14th Week	1. Team Project Beta Blog: Post-mortem analysis, announcing the contribution of each person
15th Week	1. The team summarizes the whole stage, analyzes the user data, and organizes the documents to ensure that the future team can take over.

　　The project schedule has not changed since the project is progressing smoothly.

Team Division

Team Name	Division
Yang Hao	Image recognition and text detection algorithm direction overall grasp, code integration and optimization
Zhong Wei, Shei Min	A deep learning algorithm to try and implement text detection
Li Yiju, Li Xiaoyang	Implementation and integration of single-type image recognition code
Sho Jie	Assist in the later implementation of user interface operations

Feelings

Yang Hao : Before always feel that get the project directly to start, directly think about the code how to write, now found in writing code before there are a lot of things to think about, especially now is teamwork, how to play a team of everyone's talent and enthusiasm is very important, It is also important to think about the whole idea of the project, how to make it faster and easier, avoid repeating the wheel and write the best-performing code.

Li Xiaoyang : The project has been in contact before, but because the summer has only learned a little basic knowledge, after the beginning of a long time did not touch, so basically have to learn from scratch, there are still many do not understand the place, need to study hard. Hope to get started as soon as possible, but also hope the project is progressing smoothly.

Zhong Wei : the first contact with such a team project is a big challenge for me personally. After everyone's discussion, there is a basic direction, I also actively reserve the corresponding knowledge. Although some knowledge is very difficult to understand, but I believe that the time can be understood, slowly make a point of function, in this process, their grasp of Python will continue to improve, experience will continue to enrich.

Shei Min : First contact with machine learning, it is not easy to find that machine learning is not as complex as previously imagined. While learning to practice, more and more interested in machine learning. Although the learning process encountered a lot of difficulties, but also exercise their ability to solve difficulties. Look forward to this project can bring me more experience and harvest.

Li Yiju : Team Project using python+ machine learning +OPENCV, I am mainly responsible for image processing module, although the summer vacation to OpenCV have some understanding, but there are many problems do not understand, but also encountered some of the current personal level difficult to solve problems, The next step is to study hard and strive to complete the project within the deadline.

Sho Jie : In this project, I mainly responsible for the front page of the writing, in the development, I deeply appreciate the importance of communication, and some understanding of the depth of the knowledge points to summarize, look forward to this project, can break through some technical difficulties, there are more harvest.

Second week--team project (small big guy with flying team)