In order to improve the accuracy and efficiency of existing industrial laminated thin steel-plates counting method, a thin steel-plates counting method based on Array CCD image detection was studied. An OTSU improved algorithm within rough background separation and self-adaptive segmenting threshold corrected was proposed, which was based on CLAHE (Contrast Limited Adaptive Histogram Equalization) principle, dimensional mean-filtering and morphology optimization principle of expansion and corrosion. Due to the shortcomings on missing or error peaks detecting in parameter space of traditional Hough transform, an algorithm based on auto-correlation principle and feature segmentation was proposed. Experimental results demonstrate that the proposed counting method for stacked thin steel-plates of specification 0.5 mm above is real-time, and has high statistical accuracy and good robustness.