Indian Journal of Science and Technology
Year: 2017, Volume: 10, Issue: 3, Pages: 1-8
Venkateshappa* and P. Cyril Prasanna Raj
Department of Electronics and Communication Engineering, M S Engineering College, Bengaluru - 562110, Karnataka, India; [email protected], [email protected]
*Author for correspondence
Department of Electronics and Communication Engineering, M S Engineering College, Bengaluru - 562110, Karnataka, India;
Objective: To perform image registration with sub band features computed using dual tree complex wavelets and reducing the computation time in registration. Methods/Statistical Analysis: The input images are preprocessed with Gaussian filters; features are extracted from dual tree sub bands. The dominant features are retained with feature selection process to minimize complexity in registration. Translation of images based on similar features is carried out using singular value decomposition. A refinement phase further registers the images considering similarities in features. The proposed algorithm is modeled in MATLAB, registration processes are modeled from fundamental equations. Test images are considered for validation. The test images are geometrically transformed with knows sets of translation, scale and rotational parameters. The two images are registered and the corresponding transformation parameters are compared. Critical points in the images are considered as reference points for validation. Findings: The proposed algorithm performs registration of all images with less than 10% error. The PSNR computed for all sets of data base is greater than 32dB. With dual tree sub bands, the unique features with 150 , 450 and 750 orientations are captured thus the features are twice more compared with DWT features. The feature selection algorithm considers the orientations in all three directions thus improving registration process. SVD algorithm optimizes the complexity in transformation process and refinement phase. Elimination of false features and use of appropriate sub band features has reduced the registration time by 76% thus meeting the requirement for real time applications in micro air vehicles. Application/Improvements: The improvement in computation time and registration accuracy of this algorithm is suitable for MAVs moving with speed of 20-30kmph. Multilevel decomposition will further improve the registration process, hardware accelerators can reduce computation time to few milliseconds.
Keywords: Autonomous Navigation, Dual Tree Complex Wavelets, Feature Extraction, Image Registration, SVD
Subscribe now for latest articles and news.