Skip to content

[EXPERIMENTAL] Facebook-like automatic alternative (alt) text for images using object detection with pre-trained model.

License

Notifications You must be signed in to change notification settings

ankitpokhrel/alt

Repository files navigation

This project is now available at captionai.co

CaptionAI effortlessly generates automatic image captions (alternate text), description, tags, and helps you categorize images using the power of AI.

alt  Build Status

This project is no longer maintained.

Automatic alternative (alt) text for images using object detection with pre-trained model.

Overview

A very simple PHP library to generate alternative (alt) text for images using pre-trained tensorflow model to detect objects in an image. These texts can provide context about image to visitors who are unable to see images in their browser for whatever reasons. Alt texts can also be picked up by screen readers to convert it to speech, thus, providing additional context to visually impaired people and enhancing our user experience.

This nifty little tool is inspired by facebook alt text generation process.

Example

Image may contain: cup, potted plant, laptop
Image may contain: cup, potted, laptop
Image may contain: 6 person, cup, laptop
Image may contain: 6 person, cup, laptop
Image may contain: 5 person, car, motorcycle
Image may contain: 5 person, car, motorcycle
Image may contain: 1 person, car, bus, truck, traffic light
Image may contain: 1 person, car, bus, truck, traffic light
Image may contain: chair, couch
Image may contain: chair, couch
Image may contain: pizza, dining table
Image may contain: pizza, dining table
Image may contain: dog
Image may contain: dog
Image may contain: bird
Image may contain: bird

Installation

Requires: PHP 7.1.3+, OpenCV, PHP OpenCV
Model used: SSDLite COCO v2

Pull the package via composer.

$ composer require ankitpokhrel/alt

Usage

$alt = new \Alt\Alt('/path/to/image.ext');

echo $alt->alt(); // Image may contain: 6 person, cup, laptop

$alt->setImage('/path/to/image.ext')->alt(); // Image may contain: car, motorcycle

Threshold

The classification threshold is set to 30 by default. You can adjust it as your need.

$alt->setThreshold(25);

Prefix

By default the alt is prefixed with Image may contain: . You can change this as below:

$alt->setPrefix('Image has: ');

Countable

If you want to display the count of an object, you can do so by setting countable array. person is countable by default.

$alt->setImage('/path/to/image.ext')->alt(); // Image may contain: 5 person, dog, cup

// Adding dog as countable
$alt->setCountable(['person', 'dog'])->alt(); // Image may contain: 5 person, 2 dog, cup

Default text

Default text is used if the model is not able to predict any objects within a given threshold. Default default text is No photo description available..

$alt->setDefaultText('Some scenery.');