Restoration of low-informative image for robust debris shape measurement in on-line wear debris monitoring

As a significant technique in machine condition monitoring, wear debris analysis enables investigation of machine running condition with respect to debris features including size, quantity and morphology. In particular, being capable of providing more comprehensive morphological information, three-dimensional debris features are regarded essential and often acquired through a video-based debris imaging process. However, debris images captured often suffer degradation due to debris motion blur and lubrication contamination, that hinder reliable debris features extraction. To address the image degradation issue, a new method of wear debris image restoration is developed to reduce the effect of blur. In order to avoid the expensive computation involved in blind deconvolution methods, the debris image was restored using localized boundary features. Based on the fact that debris area and background area indicate distinctive brightness, a step edge model is applied to describe the original debris boundary. Localized kernels on each side of debris are then determined. Next, restorations are conducted with the estimated kernels to produce sharper debris profiles with respect to different motion features. Final restoration is conducted by fusing the restored profiles according to the maximum local sharpness. Experimental results have demonstrated that this method allows reliable features extraction from blurred image, improving the robustness of video based wear debris analysis.