A refinement of a regional phase unwrapping technique that is driven by an integrated expert system is described. The traditional phase unwrapping algorithms used for phase shifting and Fourier transform is very noise-dependent and frequently unsatisfactory for heavily noise-ridden interferograms. THose methods that try to get rid of noises are too passive and limited. The developed method actively eliminates noises by using algorithmic as well as knowledge- based, intelligent approaches in constructing sound, not- distorted 2(pi) jump lines that divide an entire image into regions. Then regional phase unwrapping is performed region by region by adding or subtracting adequate 2(pi) multiples. The integrated expert system can correct noisy data on the iso-phase lines and the regional phase unwrapping algorithm isolates noises inside the regions without propagation.
Phase unwrapping algorithms used in phase-shift or Fourier transform techniques can be classified typically into two groups: line-based unwrapping and region-based unwrapping. Theoretically, the latter can be less error susceptible than the former. However, the pixel categorization process required in the previous region-based algorithms, utilizing the local information obtained from 3 by 3 pixels, is still error prone in many interferograms contaminated by large- scale noises. In the new approach, the developed expert system intelligently find 2 (pi) jump iso-phase lines that categorize regions having no phase jump based on global/regional information rather than the local information, that is, interferogram-specific knowledge. Then it performs phase unwrapping region by region by adding or subtracting 2 (pi) phase-wrapped band wherever region changes. The regional phase unwrapping isolates noises inherently without propagation, since every pixel's phase is unwrapped independently each other. The new algorithm is also effective especially in handling large-scaled noise- affected phase distributions.
The hybrid operation of digital image processing and a knowledge-based AI system has been recognized as a desirable approach of the automated evaluation of noise-ridden interferogram. Early noise/data reduction before phase is extracted is essential for the success of the knowledge- based processing. In this paper, new concepts of effective, interactive low-level processing operators: that is, a background-matched filter and a directional-smoothing filter, are developed and tested with transonic aerodynamic interferograms. The results indicate that these new operators have promising advantages in noise/data reduction over the conventional ones, leading success of the high-level, intelligent phase extraction.
Accurate static fringe-pattern analysis is very important for the successful application of a variety of interferometric techniques. In most cases of practical application, especially in aerodynamic flow testing, substantial noise can be introduced due to prevailing adverse environments. A means for efficiently reducing interferometric noise is thus desirable. Conventional noise reduction has mostly depended on ordinary averaging or median filtering in a squared mask to remove high-frequency components. These techniques, however, can induce some side effects of image blurring. If the structural integrity needs to be preserved, the method to be adopted should be able to eliminate noise efficiently without altering local intensity gradients, that is, local contrasts. In this paper, the concept of directional smoothing is introduced and its application to interferometric noise reduction is presented. Interferograms provide locally similar fringe directions, that is, isophase lines contaminated by noise, and thus contain directional information. In essence, the method exploits this valuable fringe directionality by setting up a slender mask of large aspect ratio along a fringe. A new value, that is, the average or median intensity of the mask, is then assigned to each pixel. The mask can be straight or curved. For a straight mask, the average direction of fringes within a processing region is employed. A curved mask is made to conform to a fringe curve. The proposed method is tested by computer simulation of experiments as well as with real interferograms. The results appear to be promising as compared with conventional techniques, especially for high-level noise.
Automation of interferogram analysis is very important for successful application of all interferometric measurement techniques. In high-speed aerodynamics or experimental mechanics, complex noise-ridden fringe patterns frequently arise due to prevailing adverse environments. In conventional practice, only local information has been heavily utilized to reduce background noise or to correct phase information. Under these circumstances, the currently available techniques, that is, fringe tracking, phase-shifting, Fourier transform, and regression methods, confront difficulties in phase unwrapping and thus need substantial interactive manual operations. The developed rule-based expert system utilizes both global/regional and local information, and makes use of expert knowledge. It can thus provide a potential for more comprehensive automation in noise reduction and phase unwrapping. The developed expert system adopts a hybrid mechanism in a single package, that is, the low-level and high-level processings to produce an optimal solution in fringe analysis. The system can be coupled with any current interferometric reduction techniques, being based on the analysis of isophase contour lines.
In this paper, the concept of directional smoothing is introduced and its application to interferogram noise reduction is presented. Interferograms provide isophase lines, that is, fringes and they thus contain directional information. In essence, the method incorporates this valuable directional information of interferograms by setting up a slender mask of relatively large aspect ratio along a fringe. The new value, that is, the average or median intensity of the pixels within the slender mask is assigned to each pixel. The slender mask could be straight or curved. For a straight slender mask, one should find the average direction of fringes within a certain chosen region. For a curved slender mask, the mask is set up along the fringe direction and may vary for each pixel. Based on computer simulation of experiments, the results appear to be promising as compared with ordinary smoothing or median filtering.