This paper introduces a technique to identify defects from fringe patterns for optical non-destructive testing and metrology. The technique relies on computation of the windowed Fourier spectrum of the fringe pattern at a given spatial frequency, and subsequent application of automated spectrum thresholding to localize the defect. The technique offers the advantages of high robustness against noise, fast implementation, high throughput and minimal operator intervention. The performance of the proposed technique is demonstrated for identifying defects of different types and sizes under varying levels of noise using numerical simulations, and practical validity is tested using experimental interferograms obtained in diffraction phase microscopy.