Please keep in mind our rule:
The more rich contrasts the image provides, the better the result of the image recognition will be!
Therefore try to follow the below guidance:
- Even though text provides many contrasts for the human eye, it has a quite negative effect on image recognition since the text normally is too small and too fragile for the used image recognition technology
- This might change according to the size of the text, so big headlines might work well
- Try to avoid especially text blocks, always try to crop your target image focusing on images, headlines or more prominent graphical elements
Avoid “empty” space
- Avoid areas in your target image which are blank white or in any other monochrome color
- Crop the image before the upload and exclude these areas if possible
- Make sure the contrast points of your image are spread homogenously over the entire image and avoid areas without any contrasts.
Avoid repetitive or “rotation synchronous” patterns
- Images like a chess board or dart board might provide a rich set of contrast but since the pattern is repetitive or “rotation synchronous” the algorithm behind the image recognition cannot recognize in which orientation the target is shown to the camera (upwards/downwards, etc.)
- as long as the repetitive pattern offers clear information for all four orientations, it should work well