Advanced visual processing techniques for latent fingerprint detection and video retargeting

An important step in an automated fingerprint identification systems (AFIS) is the process of fingerprint segmentation. While a tremendous amount of efforts has been made on plain and rolled fingerprint segmentation, latent fingerprint segmentation remains to be a challenging problem. Traditional segmentation methods fail to work properly on latent fingerprints as they are based on many assumptions that are only valid for rolled/plain fingerprints. We propose a new image decomposition scheme, called the adaptive directional total variation (ADTV) model, to achieve effective segmentation and enhancement for latent fingerprint images in this work. The proposed model is inspired by the classical total variation models, but it differentiates itself by integrating two unique features of fingerprints; namely, scale and orientation. The proposed ADTV model decomposes a latent fingerprint image into two layers: cartoon and texture. The cartoon layer contains unwanted components (e.g. structured noise) while the texture layer mainly consists of the latent fingerprint. This cartoon‐texture decomposition facilitates the process of segmentation, as the region of interest can be easily detected from the texture layer using traditional segmentation methods. The effectiveness of the proposed scheme is validated through experimental results on NIST SD27 latent fingerprint database. The proposed scheme achieves accurate segmentation and enhancement results, leading to improved feature detection and latent matching performance. ❧ In the second part, we propose two solutions for content‐aware image/video resizing (or called image/video retargeting). The first solution address the issue of texture redundancy for image retargeting. We analyze the effect of texture regularity on the performance of image resizing, and then propose an efficient texture‐aware resizing algorithm. Our solution exploits region features, including the scale and the shape information, to preserve both local and global structures. Texture redundancy is effectively reduced through texture regularity analysis and real‐time texture synthesis. The superior performance of the proposed image resizing technique is demonstrated by experimental results. Our second solution deals with conducting video retargeting on compressed‐format video data. All existing video retargeting techniques operates in the spatial pixel‐domain, which could be difficult for practical usage, as most real‐world digital videos are mainly available in compressed format. We propose a novel video retargeting system that operates directly on an intermediate representation in the compressed domain, namely, the discrete cosine transform (DCT) domain. In this way, we are able to avoid the computationally expensive process of de‐compressing, processing, and recompression. As the system uses the DCT coefficients directly for processing, only minimal decoding of video streams is necessary. Our proposed solution achieves comparable results with the state‐of‐art spatial domain video retargeting techniques, while significantly reduces the overhead of computation and storage. Though the proposed system is targeted for the latest H.264 coding standard, it can be easily applied to other video compression standards as well. ❧ Content‐aware image resizing (image retargeting) is a technique that resizes the image for optimum display on devices with different resolutions and aspect ratios. Traditional objective quality of experience (QoE) assessment methods are not applicable in the context of image retargeting because the size of the retargeted image is different from the original image. In the third part of this thesis, we start with analyzing the determining factors for humans visual perception on retargeted image quality and propose an effective quality metrics for content‐aware image resizing. The metrics extracts features from the retargeted result and uses machine learning techniques to fuse them into one single quality score. The feature design is based on three important factors: global structural distortion, local detail distortion and loss of salient information. Experimental results demonstrate that our designed metrics is more effective and general in evaluating quality of image retargeting results than existing objective metrics.