Bio-inspired algorithms for pattern recognition in audio and image processing
In this thesis, we investigate the construction of pattern recognition systems that are based on the use of novel trainable filters. The thesis addresses two important applications in the fields of intelligent audio surveillance and medical image analysis.
In the first part of the work, we propose a system for the detection of abnormal audio events, such as glass breaking, gun shots, screams, tire skidding and car accidents. The proposed system is based on CoPE filters, which make it capable to learn to detect new events by showing examples, in the same way humans learn new concepts. Such a system can be applied to improve actual surveillance systems for public or private security. We designed a deployment strategy to install microphones in places where cameras are not allowed (e.g. public toilet) or too expensive to be installed (e.g. huge parking areas). This would help the police to detect criminal acts or dangerous situations.
In the second part of the thesis, we apply COSFIRE filters to the automatic segmentation of blood vessels in retinal images. The manual analysis of retinal images is time-consuming and expensive. An automatic system can allow population screening and help doctors to recognize medical conditions such as diabetic retinopathy in an early stage, keeping at the same time low the costs of medical care.
This work provides innovative tools for pattern recognition in audio and image processing and contributes to the research trend of constructing systems for real-world application.