Quickly and Efficiently Discover Related Images via Search
This algorithm identifies features in images for computer recognition with a much higher success rate than available technologies by incorporating commonly used mathematical transforms. Part of the $5 billion computer vision market, image processing has applications in image retrieval, pattern recognition, and image compression. Image retrieval in particular relies on pattern recognition and feature extraction to find pictures similar to each other. Researchers at the University of Florida have developed an image retrieval algorithm that combines the commonly used radon and wavelet mathematical transforms. This new algorithm, called ripplet-II Transform, efficiently finds related pictures with a higher retrieval and a lower error rate than available technologies.
Algorithm that extracts features from a particular image to find related images
- Overcomes inability to accurately process edges or boundaries of images, allowing for faster and more accurate image searches
- Increases image retrieval and lowers error rate over ridgelet and wavelet transform, improving image search function
- Achieves unique representation for 2D images, allowing better feature extraction
In image processing, it is extremely difficult for computers to process 2D image features called “singularities,” sharp edges or distinct lines in images. By combining radon and wavelet transforms, this ripplet-II transform is capable of representing these singularities and using them to find similar or related images. The algorithm is also extremely efficient at classifying textures, and compensating for rotational or transformational variance.