OCR测试——字体和背景颜色

测试目的:测试图片中字体颜色和背景颜色对文字识别的影响。

一、测试图片选择

①黑色字体+白色背景;②黑色字体+橙色背景;③绿色字体+黑色背景;④绿色字体+白色背景;⑤白色字体+绿色背景;⑥白色字体+橙色背景;⑦橙色字体+白色背景;⑧混合色字体+混合色背景。

 

二、测试结果(图片)

1、黑色字体+白色背景

①v_2

ebe46d3d86fa4b2da0f49a5aaef878e1.png

召回率:100%

准确率:0.9837

耗时:0.289s

②v_3

90c1619d69ca4c3dabe32246399d943b.png

 

召回率:100%

准确率:0.9391

耗时:0.247s

③扫描王

WRITING MODEL
Essay prompt
Read the essay prompt and the brainstorm on the essay question.Classify the ideas intFor and Against by writing F or A before each statement.

 

召回率:100%

准确率:100%

耗时:

总结:v3比v2稍快一点,都能准确识别,v2比v3空格处理好,处理效果和扫描王一样。

2、黑色字体+橙色背景

①v_2

OCR测试——字体和背景颜色_第1张图片

 

召回率:100%

准确率:0.9778

耗时:0.686s

②v_3

OCR测试——字体和背景颜色_第2张图片

 

召回率:100%

准确率:0.9294

耗时:0.699s

③扫描王

UNIT ORIENTATION
Technology has impacted virtuallyevery part of our lives, and is playinga significant role in shaping today'sworld. Is it also laying the foundationfor tomorrow? Where will it take us?What will tomorrow's world be like?This unit explores some of the mostexciting advances and innovative ideasin technology that can dramaticallyinfluence our future world,andanalyzes the opportunities andchallenges they bring about.

召回率:100%

准确率:100%

耗时:

总结:v3比v2速度快一点点,都能准确识别出结果,v2比v3空格处理好,但是扫描王处理空格还是最好的。

3、绿色字体+黑色背景

①v_2

26ea2d5ac6c942e399cd829b6f9bcdb7.png

 

召回率:100%

准确率:0.9923

耗时:0.121s

②v_3

f32e5f48c10246b39b18c9de45e30935.png

 

召回率:100%

准确率:0.9782

耗时:0.144s

③扫描王

Tomorrow

召回率:100%

准确率:100%

耗时:

总结:v2比v3速度快,都能准确识别。

4、绿色字体+白色背景

①v_2

e4cf7ac38b1649bbb6232493420f52f5.png

 

召回率:100%

准确率:0.9654

耗时:0.129s

②v_3

41fcb62f54a5452c9cb5389131839dbc.png

 

召回率:100%

准确率:0.8747

耗时:0.112s

③扫描王

Academic

召回率:100%

准确率:100%

耗时:

 

总结:v3比v2速度快,只有v3识别出错了。

5、白色字体+绿色背景

①v_2

9a04898aae574c35b013dca6a93eccf3.png

 

召回率:0.97

准确率:0.9734

耗时:3.506s

②v_3

76824bc413c347468b6e31e3cc0b4830.png

 

召回率:0.82

准确率:0.9321

耗时:0.275s

③扫描王

You are going to write a short essay in response to the prompt:"Should employees speak out if their colleagues are unethical?"
To finish this task, you will first analyze a student model essay on a similar topic.Pay attentionto how the model essay is structured to respond to its essay prompt.Then you are going tolearn to use specialized vocabulary for legal terms,practice the unreal conditional in the past,and write a for-and-against essay.Finally,use these skills to write an essay about speaking outagainst unethical behavior at work on Ucampus.

 

召回率:100%

准确率:100%

耗时:

总结:准确率 扫描王>v2>v3 ;空格处理 v3最差,v2和扫描王差不多,但是v2速度很慢。

6、白色字体+橙色背景

①v_2

OCR测试——字体和背景颜色_第3张图片

 

召回率:0.9675

准确率:0.9779

耗时:0.911s

②v_3

f380e8a92ede4e1abeb759bd71e302b4.png

 

召回率:0.9712

准确率:0.9342

耗时:0.359s

③扫描王

Since the founding of the People's Republic of China,our country has made greatprogress in building a law-based government and an efficient legal system.Acase in point is the Civil Code of the People's Republic of China which came intoforce on January 1,2021.It is a major event in China's legal history,and a greatachievement in the construction of rule of law with Chinese characteristics in thenew era.As a milestone in civil legislation,it will better protect people's civil rightsand provide stronger legal support to the country's development. It is necessaryfor all students to equip themselves with enough legal knowledge in order toprotect their rights and serve the society.
 

召回率:100%

准确率:100%

耗时:

总结:空格处理  扫描王>v2>v3   时间 v2>v3   v2与v3都有一小部分没有识别到 ,扫描王全部都能识别。

7、橙色字体+白色背景

①v_2

4377267530ef48eeb5f3690b2f299e5d.png

 

召回率:100%

准确率:0.9599

耗时:0.145s

②v_3

4510a76f266042a88f61e817d9ffcbab.png

 

召回率:100%

准确率:0.9258

耗时:0.118s

③扫描王

Understanding China's Civil Code

召回率:100%

准确率:100%

耗时:

总结:三个都能准确识别  v3比v2速度上快一点,空格也都处理得很好。

8、混合色字体+混合色背景

①v_2

OCR测试——字体和背景颜色_第4张图片

 

召回率:100%

准确率:0.9627

耗时:0.980s

②v_3

OCR测试——字体和背景颜色_第5张图片

 

召回率:100%

准确率:0.9224

耗时:0.397s

③扫描王

GLOBAL PREDICTIONS FOR 2050
Computers will be
The world's population will
reach just under
THOUSANDS OF TIMES FASTER.
10 BILLIONpeople.
SOLAR POWER
will be the
world's main source of energy.
1 IN 6PEOPLE WILL
YEARS OLD.
BEOVER 65
HALF OF THE WORLD'S POPULATIONcould face severe food shortages.
Many people will live to
120 YEARS OLD.
Cars and buildings will incorporateMORE TECHNOLOGYandOPERATE INTELLIGENTLY.
Printed newspapers will
Diseases such as MALARIA will beeasilyVACCINATED against.
NO LONGER EXIST.

召回率:100%

准确率:100%

耗时:

总结:v3比v2速度快一点,三个的准确率都差不多,但是扫描王的空格处理还是要比v2和v3好点,v2和v3效果差不多。

 

整体总结:
    1)速度对比:v2和v3都略快于竞品,v3模型最快;
    2)召回率对比:竞品召回率为1,v2和v3都存在漏检现象,但是v3要严重一些;
    3)准确率对比:竞品也存在空格识别不出来的情况,空格处理上v2比v3要好很多,识别结果的精准度v2相对来说要好一些。

 

 

你可能感兴趣的:(学习)