Verification of linear barcodes print quality according to ISO/IEC15416 specifications, using Image Processing and Computer Vision techniques.
For more theoretical information, check out the following documents.
Linear-Barcodes-Verification-Project.pdf: description of the project assignment.guide-barcode-verification.pdf: in-depth description of linear barcodes print quality verification.report.pdf: description of the solution.
For an in-depth view of the functions interfaces, check out the documentation and the doc-strings inside the Python files, inside the src folder.
For a small tutorial, see the notebook usage inside the notebooks folder.
The folder tests contains the tests executed on all dataset images.
The folder out contains the excel output files for all dataset images.
Given an image containing a barcode, the task consists in verifying the print quality of the barcode, by computing some quality parameters.
Since the barcode in the input image can be rotated and can have different scales and since the input image can contain other objects apart from the barcode, the quality parameters must be computed on a standardized sub-image. We refer to this image as "refined ROI image", since it perfectly fits the Region Of Interest (i.e. the barcode) and it is refined according to some standards. More specifically, the refined ROI image is the sub-image of the input image which has the following properties.
- It contains the barcode, and the bars are perfectly vertical.
- Along the width, there are exactly
$10*X$ pixels before the first barcode bar and after the last barcode bar, where$X$ is the minimum width of a bar (in pixels). - Along the height, it perfectly fits the bar with smallest height. Basically, the height of the refined ROI image is equal to the minimum height of a barcode bar.
These images refer to the image 'UPC#01'.
Then, the print quality parameters are computed. For computing the quality parameters,
The quality parameters are computed one each scanline, by considering the scan reflectance profile, and they are the following.
- Minimum Reflectance, i.e.
$R_{\text{min}}$ . - Symbol Contrast, i.e.
$SC$ . For computing it, also the Maximum Reflectance, i.e.$R_{\text{max}}$ , is taken into account. - Minimum Edge Contrast, i.e.
$EC_{\text{min}}$ . - Modulation, i.e.
$M$ . - Defect, i.e.
$D$ . For computing it, also the Maximum Element Reflectance Non-uniformity, i.e.$ERN_{\text{max}}$ , is taken into account.
For each of these parameters, a numerical value is computed, and a symbolic grade between 'A' and 'F' is assigned, by using specific rules ('A' means very good, 'F' means very bad). In addition, a symbolic grade and a numerical value are assigned to the whole scanline.
Finally, an overall symbolic grade and an overall numerical value are assigned to the whole barcode.
Set of images containing barcodes, on which the application can be tested. Also an excel file containing the true quality parameters of these images is present. This dataset has been provided by DATALOGIC.
Different kinds of barcodes are present. In addition, there are images in which the barcode is particularly rotated or scaled. There are also images in which the barcode has the bars which are horizontally aligned instead of vertically. Let's see some examples.
There are images in which the contrast, i.e.
There are images in which the modulation, i.e.
Five images are particularly interesting for
Finally, there are images in which the defect, i.e.
Five images are particularly interesting for
Five images have
For solving our problem, a process consisting in four subsequent operations is implemented. For more information, see the report of this project. The following shown examples are about the image 'UPC#01'.
- Detect the bounding box. The bounding box surrounding the barcode in the input image is detected.
- Rotate the bounding box. The image and the bounding box are rotated such that the barcode bars are now perfectly vertical. From this operation, the ROI image is computed, which is the sub-image containing the barcode, with the bars perfectly vertical. Basically, the ROI image is the rotated image cropped around the rotated barcode. Remark: the ROI image is gray-scale.
- Refine the ROI image. The ROI image is refined, according to the standard explained before.
- Along the width, the refined ROI image is such that there are exactly 10*X pixels before the first barcode bar and after the last barcode bar.
- Along the height, the ROI image is refined in order to perfectly fit the bar with smallest height.
In order to perform this refinement, the precise and complete structure of the barcode is computed: every dimension about each bar is computed.
-
Compute the quality parameters
The quality parameters of the barcode are computed, on the refined ROI image.
As explained before, the following quality parameters are computed on each scanline:
-
$R_{\text{min}}$ , with also its symbolic grade -
$SC$ , with also its symbolic grade -
$EC_{\text{min}}$ , with also its symbolic grade -
$M$ , with also its symbolic grade -
$ERN_{\text{max}}$ , with also its symbolic grade -
$D$ , with also its symbolic grade - Symbolic grade and numerical value for the entire scanline
-
Finally, an overall symbolic grade and an overall numerical value are assigned to the whole barcode.
According to the project description, an excel output file must be generated, containing the information and results of the applied process.
In particular, the information is structured in the following sheets.
-
Global quantities. It contains the following information.
- Name of the image
- Bounding box coordinates
- Centre of the bounding box
- Angle of the rotation
-
$X$ dimension (i.e. minimum width of a barcode bar) - Height of the barcode (i.e. minimum height of a barcode bar)
- Overall symbolic grade of the barcode
-
Bars/spaces widths For each bar and space, its width is reported, in units by
$X$ dimension. It is a list, where the first element refers to the first bar, and the last element to the last bar. Basically, sequence of bars/spaces from left to right. -
Scanlines quality parameters For each scanline, its quality parameters are reported. Namely:
-
$R_{\text{min}}$ and its symbolic grade -
$SC$ and its symbolic grade -
$EC_{\text{min}}$ and its symbolic grade -
$M$ and its symbolic grade -
$D$ and its symbolic grade - Symbolic grade and numerical value
-
Actually, this is the basic output format. The user can specify to build a richer output file, containing more information.
The most important functions are now briefly introduced. Each of them is contained in a separate Python file, since they can involve several auxiliar functions, with several lines of code.
Function verify_barcode. This is the main function, which performs the overall process of barcode verification. So, it performs all the four recquired steps. Basically, it consists in the subsequent application of the four functions detect_boundingBox, rotate_boundingBox, refine_ROIimage, compute_quality_parameters.
Optionally, the user can get information about the execution time of each operation. Furthermore, he can also get plots for visualizing the results of each operation.
The following functions are more specific and lower-level functions, for performing the recquired sub-tasks
Function detect_boundingBox. It detects the bounding box surrounding the barcode.
Function rotate_boundingBox. It rotates the input image and the bounding box such that the bars become perfectly vertical.
Function refine_ROIimage. It refines the ROI image, according to the recquired standard format. In order to do so, the complete barcode structure is computed.
Function compute_quality_parameters. It computes the quality parameters on the barcode.
Function build_output_file. It build the output file.
Example of application of the verify_barcode function. It returns four dictionaries, containing the information and results of the four operations. Optionally, the user is asking to print the timing information and to make two specific plots.
# Verify the barcode print quality of the given input image
image_path = './dataset/UPC#01.bmp'
visualization_dict = {
'visualize_refinedRoi': True,
'visualize_scanlines_qualityParameters':True
}
detection_dict, rotation_dict, refinement_dict, overall_quality_parameters_dict = verify_barcode(image_path, visualization_dict=visualization_dict, verbose_timing=False, create_output_file=True)The following is the obtained timing information.
TIMING INFORMATION (milliseconds)
Detect bounding box: 0.012889862060546875
Rotate bounding box: 0.01150059700012207
Refine ROI image: 0.04480409622192383
Compute quality parameters: 0.23713088035583496The following are the plots.
.
├── dataset # Input images and excel file containing the true quality parameters
├── images # Images shown in this README or in the notebooks
├── notebooks # Folder containing a small tutorial notebook
├── out # Folder containing the excel output file for all the dataset images
├── src
│ ├── build_output_file.py
│ ├── compute_quality_parameters.py
│ ├── detect_boundingBox.py
│ ├── refine_ROIimage.py
│ └── rotate_boundingBox.py
│ └── utils.py
│ └── verify_barcode.py
├── tests # Folder containing the tests on all datasets images
├── .gitignore
├── guide-barcode-verification.pdf
├── LICENSE
├── Linear-Barcodes-Verification-Project.pdf
├── report.pdf
└── README.md
This project is licensed under the MIT License - see the LICENSE file for details.