Abstract: Object detection and recognition is actively studied in computer vision and machine learning. Particularly, in recently years, topic models such as latent Dirichlet allocation (LDA) has achieved great success in unsupervised recognition and localization of objects. However, LDA ignores the spatial relationships among image regions. To address this issue, conditional random field (CRF) introduces local dependence to improve the classification accuracy of image patches. In this paper, we propose a latent Dirichlet allocation-conditional random field (LDA-CRF) model by combining LDA with CRF. CRF is trained with topic features generated by LDA, while LDA generates topic information by utilizing structured class labels provided by CRF. Experimental results show that LDA-CRF performs better than CRF in object detection and recognition.