Product indexing method and system thereof

A product indexing method comprising:

• receiving an input image;
• analyzing the input image, the analyzing including:
• identifying a product displayed in the input image;
• selecting a product image from the input image, the selected product image being an image of the identified product displayed in the input image;
• selecting, based on the selected product image, a context image from the input image, the selected context image being one or more sections displayed in the input image other than the product image that provide contextual information for the identified product displayed in the selected product image;
• predicting, based on the identified product, one or more product categories for the identified product from among a plurality of pre-defined product categories, each pre-defined product category covering one or more products, each pre-defined product category having one or more required viewpoints for the one or more covered products in the pre-defined product category, wherein each of the one or more predicted product categories include one or more required viewpoints for the identified product, wherein the predicting of each product category includes:
• generating a product prediction score, the product prediction score representing a probability that the identified product belongs to the predicted product category, the product prediction score including:
• an image prediction score, the image prediction score representing a probability that the selected product image belongs to the predicted product category;
• and
• a text prediction score, the text prediction score representing a probability that one or more textual information identified in the input image belongs to the predicted product category;
• selecting, from among the one or more predicted product categories, a selected predicted product category, the selected predicted product category being the predicted product category having the highest product prediction score;
• verifying the selected product image to determine whether the selected product image needs to be refined, the verifying based on the selected context image and the predicted product category;
• determine, based on the verifying, whether the selected product image needs to be refined;
• responsive to a determination that the selected product image does not need to be refined:
• identifying a viewpoint of the identified product in the selected product image, the viewpoint of the identified product in the selected product image identified based on a comparison of the selected product image with one or more of the required viewpoints in the predicted product category;
• extracting the selected product image based on the identified viewpoint of the identified product in the selected product image;
• and
• indexing the extracted product image and the selected predicted product category.