Recent quantum leap in far-field laser techniques has advanced noncontact implementation of nondestructive ultrasonic imaging, in pursuit of enhanced accessibility, detectability and practicability. Nevertheless, when laser-generated thermoelastic waves are extended to thick waveguides, they manifest fairly low signal-to-noise ratios (SNRs), along with severe wave diffusion, consequently lowering image resolution and contrast. With these motivations, a laser-ultrasonics imaging approach is developed, in conjunction with i) entropy-polarized bilateral filtering (Entropy-P-BF) for signal denoising, and ii) minimum variance (MV) beamforming for defect imaging, targeting at precise characterization of a submillimeter defect (with its characteristic dimension being smaller than the wave diffraction limit) in a thick waveguide. The entropypolarized bilateral filtering denoises laser-induced ultrasonic wave signals via a two-dimensional convolution, the weight matrices of which are continuously updated according to local noise and uncertainty. With an elevated SNR, MV beamforming subsequently conducts an apodized beamforming to image the defect. Experimental validation is conducted by imaging a void-type defect, 0.7 mm only in its diameter, in a jet aero-engine turbine disk. Results prove that the developed approach is capable of characterizing a submillimeter defect accurately in a thick waveguide with thickness similar to 25 times the wavelength of laser-induced shear wave, regardless of a fairly low SNR (<1dB).