How it works
1. Object Recognition
Individual pixels of the image are analysed and their connectivty to their nearby pixels is recognised. This enables the object recognition model to know if two or more pixels are part of the same object.
2. Object Classification
Every recognised object has certain characteristics which hold true for all similar objects and can thus be classified into a category.
For instance, all basketballs will be spherical in nature and have well defined markings on them. Another example is that of a face. Every human face will have certain data points which are uniform across all people. The distance/connectity between these points can vary but they retain a certain ratio. If this ratio is maintained for an object, then we can classify that object as a face.
If the object does not match any existing object, then it is treated as a new category and once similar objects are found, is added as a verfied category.
- Monitoring and filtering adult/violent content: Any social platform which allows users to upload images may need to hide the image behind a disclaimer that it is meant only for adult audiences. Another platform may pose a restriction on images which might be advertisements. Such uploads can be prevented.
- Ensuring uniformity for photographs for visas: While applying for a country's visa, there are specific conditions which the photograph must satisfy. For instance, wearing any form of headgear might not be allowed. If we're able to flag images and inform the user to upload a different image, it will reduce the number of re-applications needed due to improper images.
- Content Curation: Social Media platforms try ot bring together thier users by introducing them to peolpe who have similar interests. Their images can have the required tags automatically and will be able to view similar tag possesing images. This will also help cut down on cases where the user adds an incorrect tag for the image therby hampering any training algorithm or creating dissonance.
- Inventory Management: The number of similar objects in a picture can be counted to assess the stock for a particular product. This can be used to get a rough estimate of the number of articles present.
- Crop Health Monitor: Onset of diseased crops can be recognised and remedies for the same can be given out for farmers. For this, a model will be trained based on images of healthy and diseased crops and will be able to clasify the disease in it's different stages. A predictive model will help ensure proper medication can be applied to such crops before thet are rendered unusable.