Image Processing functionality analyses images that can be found in public articles or posts and their reproductions, and classifies them to pre-defined labels. The outcome is the overall image’s scene categorisation, and works with an expected accuracy of over 80%.
Brand image recognition
Brand image recognition (& video) feature will result in the recognition of branding on elements of an image, such as objects in the foreground or background. The key way of working will be through accurate logo detection.
Age and gender detection will also an outcome of the image analysis.
Structure content analysis
The questions to answer and the respective labels for image classification are the following:
WHO was involved? Information around demographics, relation of people in the picture (i.e. family, co-workers…), celebrities, etc
WHAT happened? what objects are present in the picture (fashion items, beauty items, food etc)
WHEN did it take place? events (i.e. music concert, sports event…)
WHEREdid it take place? i.e. (INDOOR (public indoors vs. private indoors) HOUSE AREAS (kitchen, rooms, bathrooms, living rooms) / OUTDOOR – SCENERY: (beach, nature, city, public areas- restaurants, cinema, theatre, bars etc, football fields, stadiums etc) / TRAVEL MEDIUM (airplanes- airports – boats – bike – cars – trains – bus)
WHY did that happen?
HOW did it happen? i.e. OCCASIONS: socializing, coffee occasions, drinking occasions, dancing occasions, chilling, eating occasions, cooking occasions, entertainment, family time, sports, outdoor activities, meditating/ praying, touristic attractions, pampering (beauty – haircut – spa) – FACIAL EXPRESSIONS EMOTIONS RECOGNITION
Features and Modules
The purpose of the Logo Detection module is to identify known logo(s) within a given set of input images. This module should be trained with a restricted set of sample images and run autonomously and potentially in a parallel manner.
The Scene/Object Recognition module will be responsible for categorizing input images based on the following axes:
- persons: demographics, relation of people in the picture (i.e. family, co-workers), celebrities.
- objects: i.e. fashion items, beauty items, food.
- events: i.e. music concert, sports event
- places: Indicate classifications are the following:
- INDOOR: public indoors vs. private indoors)
- HOUSE AREAS: kitchen, rooms, bathrooms, living rooms
- OUTDOOR – SCENERY: beach, nature, city, public areas, restaurants, cinema, theater, bars, football fields, etc
- TRAVEL MEDIUM: airplanes, airports, boats, bikes, cars, trains, buses.
The purpose of the Face Recognition / Demographics module is to be able to identify distinct faces and extract gender and demographics-related information, the latter being defined within certain age periods, indicatively:
The Image Ranking module is responsible to produce an ordered list of images by relativity to a given cluster, which comprises a set of logically similar web or social media articles, in terms of topic and/or categorization.