Democritus University of Thrace (DUTH)
Σχεδιασμός και υλοποίηση ηλεκτρονικών κυκλωμάτων πραγματικού χρόνου για όραση μηχανής
Abstract
dc:descriptionImage scaling schemes are becoming more popular due to the wide use of digital imaging devices and demanding due to the increasing sensor capturing capabilities in spatial resolution. Various applications require image zooming, such as digital cameras, electronic publishing, 3G mobile phones, medical imaging and image processing. An image scaling technique performs spatial interpolation since it enlarges the spatial resolution of the input image and completes the missing data from the spatially adjacent data using some type of interpolation operation. Furthermore, it tries to preserve the spectral contents of the input image. In this PhD Thesis, we propose the design and implementation of an interpolation scheme for performing image scaling by utilizing a dynamic mask combined with a sophisticated neighborhood averaging fuzzy algorithm. The functions that contribute to the final interpolated image are the areas of the input pixels, overlapped by a dynamic mask and the difference in intensity between the input pixels. Fuzzy if-then rules for these two functions are presented in order to carry out the interpolation task. Simulation results have shown a fine high-frequency response and low interpolation error in comparison to other widely used algorithms. The interpolation can be applied to both gray-scale and color images, for any scaling factor. The proposed hardware structure is implemented in a field programmable gate array (FPGA) chip and is based on a sequence of pipeline stages and parallel processing, in order to minimize computational times. The fuzzy image interpolation implementation, combines a fuzzy inference system and an image interpolation technique in one hardware system. Its main features are the ability of accurate approximation of the Gaussian membership functions, used by the fuzzy inference system, with very few memory requirements and its high frequency performance of 65 MHz, making it appropriate for real-time imaging applications. The system can magnify gray-scale images of up to 10-bit pixel depth. The maximum input image size is 1024 x 1024 pixels for a maximum of 800% Furthermore, a bio-inspired interpolation algorithm suitable for image scaling is also proposed. A log-polar neighbor model is adopted, utilizing the feature of applying larger weights to pixels at the center of the interpolation region and logarithmically decreasing weights to pixels away from the center. The interpolation is performed in the Cartesian plane without requiring the full transformation of the image to the log-polar plane. Experiments show that in both visual comparisons and quantitative analysis, the results extracted by the proposed log-polar neighbor model are better than those extracted from pixel repetition, bilinear and bicubic interpolation. Due to the unique characteristics and wide applications of image scaling, a separate study of their evaluation methods was performed in this PhD Thesis. We studied different existing methods for image scaling algorithm evaluation. In addition, one new evaluation method is proposed which is directly related to ground truth optically scaled images. Digital image scaling is classified into the general interpolation scheme. Two main classes of criteria are used as the basis for the assessments: The fidelity criteria and the computational complexity criteria. All methods are briefly described and comparative discussions are provided. This survey is helpful for an appropriate use of existing evaluation methods and for improving their performance as well as for systematically designing new evaluation methods. Digital image stabilization schemes are becoming very popular and demanding due to the wide use of digital imaging devices. The image stabilization process removes or compensates the unwanted motion occurred when capturing the image sequence. Many image processing applications require stabilized sequences for input while other present substantially better performance when processing stabilized sequences. Here, we propose a novel digital image stabilization scheme based on Independent Component Analysis (ICA). The method utilizes ICA and information obtained from the image sequence in order to deconvolve the ego-motion from the unwanted motion of the sequence. We notice that the motion observed in image sequences captured from consumer electronics, such as handheld cameras and 3G mobile phones, is mainly caused by two independent motions; the camera motion (ego-motion) and the undesired hand jitter (high-frequency motion). The independence property of these two primary motions facilitates the application of the ICA method for deconvolution. Sets of estimated local motion vectors of the sequence are introduced to the ICA system for separation. Then, we process the unmixed motion vectors in order to classify the signals into ego-motion and high-frequency motion. Subsequently, when the permutation ambiguity is resolved, the appropriate sign and energy are assigned to the ego-motion vector, resulting to the stabilized image sequence. Experimental results have shown that apart from the successful deconvolution of the two different motions, the proposed scheme exhibit superior performance compared to other digital image stabilization algorithms. The two above features, zooming ability and the video stabilization, are considered as a de-facto in recent consumer imaging devices. Both characteristics can be found in a single embedded system of a digital still camera, video recorder or even in a 3G mobile phone. These two features can be enabled at the same time in a single imaging device, producing a zoomed and stabilized sequence. However, in these operation conditions, the user can notice in the output video a latency and an undesired shaking, which aggregates while the user increases the magnification factor. This PhD Thesis presents a novel architecture for integrating digital stabilizer with the camera zooming process. The proposed architecture scheme allows integration for both optical or digital zooming operation scenarios. The stabilizer adjusts its operational parameters from the current optical zooming status for a refined local motion estimation. Subsequently, the global motion compensation vectors, produced by the digital stabilizer, are sent to the digital zooming system, where the image compensation is merged within the interpolation process. Experimental results indicate that the proposed architecture can improve not only the quantitative performance of digital stabilization but also the computational efficiency when zooming is employed. Summarizing, in the present Thesis, techniques for digital image scaling and digital image stabilization are presented in order to improve their performance in terms of accuracy and speed. Furthermore, a novel architecture for integrating digital image stabilization with digital image scaling is also proposed, improving both the quantitative performance and computational efficiency of the whole process.
Degree
thesis:*- Grantor dc:publisher
- Democritus University of Thrace (DUTH)
- Year dc:date
- 2009
Author and committee
dc:creator, dc:contributor.*- Authors dc:creator
-
- Amanatiadis, Angelos
- Αμανατιάδης, Άγγελος
Subjects
dc:subject × 9- Όραση μηχανών
- Πραγματικός χρόνος
- Ψηφιακή κλιμάκωση εικόνας
- Ψηφιακή σταθεροποίηση εικόνας
- Ενσωματωμένη αρχιτεκτονική δυναμικής σταθεροποίησης και κλιμάκωσης εικόνας
- Επιστήμες Μηχανικού και Τεχνολογία
- Επιστήμη Ηλεκτρολόγου Μηχανικού, Ηλεκτρονικού Μηχανικού, Μηχανικού Η/Υ
- Engineering and Technology
- Electrical Engineering, Electronic Engineering, Information Engineering
Rights
- Language dc:language
- gre
Identifiers
dc:identifier.*- Identifier
- 10.12681/eadd/20957