As a vital part for industrial automatic monitoring systems, this paper presents the algorithm for top window recognition on screen-grabbed images. For better performance in terms of accuracy, efficiency, and robustness, we introduce a novel, background-based image feature of window region, i.e., locally-maximum uniform blocks extracted from regions of uniform color, to represent the practically unique spatial structure of background regions in different windows. The recognition procedure consists of several rather simple steps, c.f., foreground-based methods, which include (i) edge extraction, (ii) uniform block extraction, (iii) uniform block matching. The template window with highest matching score is then identified the top window, with its location determined simultaneously to facilitate further analysis of window content. Experimental results indicate that the proposed approach can achieve perfect recognition results for realistic image data from the industry, and also fulfill the near real-time processing requirement of on-line monitoring tasks.