A fundamental component of simultaneous localization and mapping systems is loop closure detection. For consistent mapping, accurate loop closure detection is crucial to reduce the drift of the estimated trajectory. As the map size increases, loop closure detection performance becomes more critical, but it gets harder and needs more computational time to find correct loop closure candidates. This paper presents an extension to a state-of-the-art RGB-D SLAM system to increase accuracy of large-scale mapping in real-time. The proposed extension uses a straightforward visual place recognition method to determine loop closure candidates. The method combines global and local image features through employing image histograms and keypoint matching. Four different place recognition techniques composed of complementary steps of the method are studied: histogram only, brute-force keypoint matching, hierarchical clustering, and adaptive thresholding. The extended RGB-D SLAM system is assessed on a popular dataset in terms of accuracy and speed. The quantitative results show that the proposed method improves accuracy up to approximate to 42% and works fast enough to meet real-time requirements. The method enables to perform real-time large-scale indoor mapping effectively on CPU.