当前位置:网站首页>How to optimize the decoding performance of dynamsoft barcode reader

How to optimize the decoding performance of dynamsoft barcode reader

2020-11-06 22:28:25 roffey

Dynamsoft Barcode Reader SDK A multifunctional barcode reading control , It only needs a few lines of code to embed the barcode reading function into Web Or desktop applications . This can save months of development time and cost . It can support multiple image file formats as well as those obtained from cameras or scanners DIB Format . Use Dynamsoft Barcode Reader SDK, You can create powerful and practical barcode scanner software , To meet your business needs .

Click to download Dynamsoft Barcode Reader The latest version

Many companies like to use Dynamsoft Barcode Reader SDK, Because it has flexible parameter configuration and powerful decoding ability for multiple barcodes . In this paper , Let's take a look at the bar code SDK Templates and possible ways to optimize decoding performance from a developer's point of view .

How to configure the template for decoding performance

If you've never tried Dynamsoft Barcode Reader SDK, You can play in the online bar code playground , Just change the mode to directly compare performance differences .
条码解码性能的模板
Besides , If you're an expert , You can click Advanced settings to adjust a series of parameters .

For the convenience of developers , I asked Github Five useful template files have been uploaded :

 

  • Speed.json
  • Balanced.json
  • Coverage.json
  • Morecoverage.json
  • Mostcoverage.json

Decoding speed or decoding accuracy ? You can weigh the trade-offs , It depends on the specific usage scheme .

 

This is my test image :
所有条形码类型
Let's look at the detection accuracy and time cost of using different templates :

 

BarcodeReader.exe AllSupportedBarcodeTypes.png  license.txt  speed.json   
Total barcode(s) found: 12. Time cost: 63 ms

BarcodeReader.exe AllSupportedBarcodeTypes.png  license.txt   balanced.json
Total barcode(s) found: 13. Time cost: 140 ms
 
BarcodeReader.exe AllSupportedBarcodeTypes.png  license.txt   coverage.json
Total barcode(s) found: 13. Time cost: 844 ms
 
BarcodeReader.exe AllSupportedBarcodeTypes.png  license.txt   morecoverage.json
Total barcode(s) found: 13. Time cost: 1610 ms
 
BarcodeReader.exe AllSupportedBarcodeTypes.png  license.txt   mostcoverage.json
Total barcode(s) found: 13. Time cost: 3156 ms

As far as I'm concerned , To ensure the accuracy and decoding speed , The most appropriate template is balance.json.

 

Can multithreading speed up the performance of multi barcode decoding ?

According to our common sense , The time cost of decoding a single barcode should be less than that of decoding multiple barcodes . therefore , One possible optimization for reading multiple barcodes is to create multiple worker threads , To process different barcode symbols at the same time .

This is the code used to decode one-dimensional and two-dimensional barcodes in sequence :

 

barcode_decoding(buffer, size, BF_CODE_39, 1, license, config);
barcode_decoding(buffer, size, BF_QR_CODE, 1, license, config);
barcode_decoding(buffer, size, BF_PDF417, 1, license, config);
barcode_decoding(buffer, size, BF_DATAMATRIX, 1, license, config);

The total time cost is 407 millisecond :

Thread id: 22536. Type: CODE_39
Thread id: 22536. Total barcode(s) found: 1. Time cost: 235 ms
 
Thread id: 22536. Type: QR_CODE
Thread id: 22536. Total barcode(s) found: 1. Time cost: 47 ms
 
Thread id: 22536. Type: PDF417
Thread id: 22536. Total barcode(s) found: 1. Time cost: 62 ms
 
Thread id: 22536. Type: DATAMATRIX
Thread id: 22536. Total barcode(s) found: 1. Time cost: 63 ms

In order to optimize decoding performance , I can create four threads to perform the same operation :

 

 

int starttime = gettime();
thread t1(barcode_decoding, buffer, size, BF_CODE_39, 1, license, config); 
thread t2(barcode_decoding, buffer, size, BF_QR_CODE, 1, license, config);
thread t3(barcode_decoding, buffer, size, BF_PDF417, 1, license, config);
thread t4(barcode_decoding, buffer, size, BF_DATAMATRIX, 1, license, config);
t1.join();
t2.join();
t3.join();
t4.join();
int endtime = gettime();
printf("Thread time cost: %d ms\n\n", (endtime - starttime));

The final time cost is 265 millisecond :

 

 

Thread id: 24024. Type: QR_CODE
Thread id: 24024. Total barcode(s) found: 1. Time cost: 78 ms
 
Thread id: 17384. Type: DATAMATRIX
Thread id: 17384. Total barcode(s) found: 1. Time cost: 78 ms
 
Thread id: 24264. Type: PDF417
Thread id: 24264. Total barcode(s) found: 1. Time cost: 94 ms
 
Thread id: 4060. Type: CODE_39
Thread id: 4060. Total barcode(s) found: 1. Time cost: 265 ms
 
Thread time cost: 265 ms

up to now , It seems pretty good . however , If you pass multiple barcode types to Dynamsoft Barcode decoding API, Then something magical will happen :

 

 

barcode_decoding(buffer, size, BF_CODE_39 | BF_DATAMATRIX | BF_QR_CODE | BF_PDF417, 1, license, config);

It's faster than your own multithreading solution :

Thread id: 20308. Type: PDF417
Thread id: 20308. Type: QR_CODE
Thread id: 20308. Type: DATAMATRIX
Thread id: 20308. Type: CODE_39
Thread id: 20308. Total barcode(s) found: 4. Time cost: 250 ms

The reason is that all of Dynamsoft Barcode decoding API Both are implemented in threads . therefore , You don't need to create threads to optimize decoding performance .

How does the number of threads affect Dynamsoft Barcode SDK performance ?

 

You may have noticed , There is one named maxAlgorithmThreadCount Parameters of . We can improve by increasing the number of threads SDK Performance ?

I did a simple test based on the hardware thread :

 

const auto processor_count = std::thread::hardware_concurrency();
int minimum_count = 1, minimum_timecost = 0;
    for (int i = 0; i < processor_count; i++) 
    {
        printf("Thread count: %d. ", i + 1);
        int timecost = barcode_decoding(buffer, size, formats, i, license, config);
        if (i == 0) 
        {
            minimum_count = 1; 
            if (timecost > 0)
            {
                minimum_timecost = timecost; 
            }
        }
        else {
            if (timecost < minimum_timecost)
            {
                minimum_count = i + 1;
                minimum_timecost = timecost; 
            }
        }
    }
    printf("Multi-thread best performance: thread_count = %d, timecost = %d \n\n", minimum_count, minimum_timecost);

Every time I run the app , We'll get different results . By using my test images , There is no significant difference in performance :

 

Thread count: 1. Thread id: 26376. Total barcode(s) found: 13. Time cost: 141 ms
 
Thread count: 2. Thread id: 26376. Total barcode(s) found: 13. Time cost: 141 ms
 
Thread count: 3. Thread id: 26376. Total barcode(s) found: 13. Time cost: 125 ms
 
Thread count: 4. Thread id: 26376. Total barcode(s) found: 13. Time cost: 140 ms
 
Thread count: 5. Thread id: 26376. Total barcode(s) found: 13. Time cost: 157 ms
 
Thread count: 6. Thread id: 26376. Total barcode(s) found: 13. Time cost: 203 ms
 
Thread count: 7. Thread id: 26376. Total barcode(s) found: 13. Time cost: 156 ms
 
Thread count: 8. Thread id: 26376. Total barcode(s) found: 13. Time cost: 140 ms
 
Multi-thread best performance: thread_count = 3, timecost = 125

 

obviously , A test image doesn't make any sense . Ideally , You should use image datasets to measure performance . therefore , If you are interested , Do it now .

This article is reprinted from 【 Huidu technology 】. Any reprint is welcome , But be sure to indicate the source 、 Do not modify the original link , Respect for the work of others

版权声明
本文为[roffey]所创,转载请带上原文链接,感谢