A good image target, that is, an image that the application will recognize fast and without problems, is an image that has rich detail, good contrast, asymmetry and does not have repetitive patterns. Overall, photos work well as targets.
The recognition that the Magipix application makes look for very distinct and contrasted regions in the image.
Blurred images and color gradients, because they have no contrast or detail, do not generate good recognition.possuírem contraste nem detalhes, não geram um bom reconhecimento.
In the example below, the sky, which occupies most of the image, has a gradient color
transition and has almost no contrast. Such regions of the photo will not generate good recognition.
In the example below, the image itself being rich in detail, is very blurred and makes application recognition difficult.
Ideally, the image has regions with the most striking color and luminosity variations.
In the example below, the photo shows regions with marked and defined color variations. For the application it is easy to recognize regions with this type of textures.
Our app recognizes images at any angle, so if you aim at an image that is tilted or even upside down it will be able to recognize the image, but symmetrical images do not favor recognition.
In the example below we see the similarity between two different regions of the same image. In this image the formats are repeated, then several regions of the image are identical or very similar, making recognition difficult.
In the example below we see an area of the image and that same area rotated by 90 degrees. In this case in addition to the pattern being repeated, if a pattern module is rotated at 90 or 180 degrees it will remain the same, so it becomes difficult to set the angle of the image and this makes recognition difficult.
Circular or rounded shapes usually have a similar problem, their edges at different angles end up being indistinguishable.
In the example below it is possible to notice that the cutout B when rotated is identical to the cut out A.
The images that generate the best recognition are those that have asymmetric and contrasting shapes. This doesn't mean that its image can't contain sinuous and less defined forms, but it is important that there are contrasting elements distributed in the image so that there is a good recognition.
In the following examples it's possible to see that the images have very distinct regions between them and that if you rotate a piece of the image it becomes different from the unrotated piece, that is, they are images with good points of recognition.