Computer vision, image processing and machine learning
Computer vision, image processing and machine learning
Image: Colourbox.com
Empowering computers to interpret and comprehend image content can revolutionize numerous sectors. For instance, distinguishing a malignant tumour from a benign one in medical imaging could significantly enhance diagnostic outcomes. Similarly, augmenting an autonomous vehicle’s ability to discern lanes, traffic signs, and pedestrians from camera feeds can lead to safer and more efficient transportation. While substantial progress has been made in this field, it’s not without its challenges.
Understanding decisions made by AI
To fully harness the potential of Artificial Intelligence (AI), it’s crucial that we trust its decisions. This trust stems from our understanding of how and why AI reaches its conclusions. In light of this, researchers at Colourlab have been working on enhancing the interpretability of decisions made by deep learning models. They’ve developed a novel method, Stabilized LIME for Consistent Explanations (SLICE), which offers consistent explanations for any deep learning model (Bora et. al. 2024).
Predicting image quality
Facial recognition systems, already prevalent in various applications such as access control and user authentication, are another area of focus. The quality of images captured can be influenced by numerous factors, including lighting conditions and facial expressions. Poor image quality can compromise system performance and potentially grant access to the wrong individual. Predicting image quality is therefore vital. This is an area where Colourlab researchers, in collaboration with their partners, have been focusing their efforts and their results show that their proposed solution produces meaningful pixel-level qualities, enhancing the interpretability of the face image and its quality (Terhörst et al 2024).
These are just a few examples of the exciting research Colourlab is conducting in the realms of computer vision, image processing, and machine learning. For more detailed insights, please refer to the linked research papers below.
This text is partially generated by Microsoft Copilot (AI)