Image saliency detection
Identify the most important objects in an image
To improve personalization in its service offerings, a global leader in marketing automation asked us to create a way to identify the most salient visual elements, or focal points, within images across thousands of websites.
Machine learning helped us identify the most important objects.
To automatically identify the most relevant or salient visual elements within thousands of images, we used a process called image saliency, which enables its user to detect logos, branding, and other key design elements in websites. Image saliency detection can be used by marketers in AI-driven design automation to personalize marketing activities or to provide automated design services.
- Research & development
- Apply a pre-trained neural network model
- Develop neural network model
- Create and train convolutional neural network (CNN)
- Crop an image without losing the main crux of the image
- Deliver trained neural network model
The saliency map above shows how a deep convolutional neural network (CNN) detects salient objects in an image. The first image is the “input.” The second image is the “output.”
Image saliency analysis
o analyze image saliency in each image, we applied a pre-trained machine learning model that could help us identify saliency data, brightness, ratio, and cropping guidelines.
We used guiding questions, such as:
- Can I put text or other assets on this part of the image?
- What is the primary focal point of the image?
- What is the ratio of this image?
- What is the brightness of the top section?
Each image was analyzed using a deep convolutional network trained for saliency prediction. The resulting saliency data was then split into four quadrants, which were analyzed independently to provide a saliency rating from 0 to 1, where 1 is highly salient and 0 is low saliency. Lower saliency quadrants generally are better suited for asset overlays such as headings or logos.
We also calculated primary and secondary focal points. Focal points can be used to influence overlay positions and provide image cropping guidelines.
Client outcome – a machine learning model
Our client can now apply a machine learning model to detect logos, branding, and other key design elements within millions of websites to automate its design services and personalize its marketing activities.
Get started with machine learning today.
next case study
innovation research, machine learning, media / technology