Image Quality Assessment
Document Reader SDK may extract information incorrectly from identity documents for several reasons.
The quality of the device’s camera is one significant factor. Higher resolutions can capture more detail, improving the accuracy of OCR. Conversely, lower resolutions may miss details, leading to OCR errors. If the camera fails to focus properly on the document, it will be inaccurately recognized.
The presence of glares obscure parts of the document, making the scanning process more challenging. Similarly, blur, can impede ability to accurately recognize the document's information.
The perspective angle from which the document is scanned also affects the outcome. If the document is not scanned from a straight-on angle, Document Reader SDK may struggle to recognize skewed information correctly. Moreover, if the document's edges are not fully within the camera's view, the document will be inaccurately recognized.
The mentioned factors are analyzed within Image Quality Assessment that includes:
| Feature | Mobile SDK | Web Service | Web Components | Hardware devices |
|---|---|---|---|---|
| Resolution |  |
|
|
|
| Focus | |
|
|
|
| Glares | |
|
|
|
| Perspective angle | |
|
|
|
| Bounds | |
|
|
|
| Portrait presence | |
|
|
|
| Colorness | |
|
 |
|
Let's explore each image quality check in detail.
Resolution
The "resolution" image quality check ensures that the image of the document meets the required resolution. The image is analyzed to determine its pixel density, typically measured in DPI (dots per inch) or PPI (pixels per inch). The resolution is compared against the predefined value. The output provides a result: "resolution enough," "resolution not enough", or "check not performed" if the check is skipped.
Focus
The "focus" image quality check ensures that the image of the document is sharp and free of blurriness. The image is analyzed using algorithms that assess sharpness, often by evaluating edge clarity and contrast. The output provides a result: "focus enough," "focus not enough", or "check not performed" if the check is skipped.
Glares
The "glares" image quality check ensures that the image of the document is free from reflective glare, which can obscure important details. The image is analyzed using algorithms that detect glare by identifying areas of high brightness and low contrast. The output can indicate three states: "glares absent," "glares present", or "check not performed" if the check is skipped.
Perspective angle
The "perspective angle" image quality check ensures that the document image is captured from a proper angle, minimizing distortions caused by perspective. Algorithms are employed to analyze the image and detect any perspective distortions, typically by examining the alignment of document edges and corners. The angle of the document is compared against the predefined value. The output can indicate three states: "correct perspective angle," "incorrect perspective angle", or "check not performed" if the check is skipped.
Bounds
The "bounds" image quality check ensures that the entire document is captured within the image frame without any cropping or truncation. This process involves analyzing the document image to detect if any edges or corners of the document are cut off or obscured. The output can indicate three states: "bounds complete," "bounds incomplete", or "check not performed" if the check is skipped.
Portrait presence
The "portrait presence" image quality check verifies that a portrait image is present on a given identity document. Face detection algorithms are used to detect the presence of a face. The output provides a result: "portrait present", "portrait not present", or "check not performed" if the check is skipped.
Explore Advanced Portrait Authenticity Check
Colorness
The "colorness" image quality check determines whether the image is black-and-white or contains color information. This process involves analyzing the color composition of the image pixels to detect if there are variations beyond grayscale. The image is classified as black-and-white if it contains grayscale tones. The output provides a result: "colored", "not colored", or "check not performed" if the check is skipped.
Image Quality Requirements
Explore Image Quality Requirements to increase the recognition accuracy.