Construction And Experimental Analysis Of House-Tree-Person Image Data Set

Abstract

House-tree-person test (HTP) is a projected mental health test based on images. The image data set of the HTP is the basis of intelligent detection and analysis of HTP images. Although HTP is widely used, there are few general data sets. To this end, this study collected images in three ways: Internet download, book scanning, and reality drawing, obtaining thousands of digital images of the HTP. Then, the house, tree and person elements in each image were segmented to form four image data sets of the integral, house, tree and person. Therefore, according to the rules, the elements and features of the house, tree and person are individually calibrated, and the experimental data are divided. Finally, the YOLO target- detection model was used for the experiment, and the detection precision of the four data sets under different quantities was analyzed and compared. The experimental results showed that the average detection precision were equal to 0.828, 0.897, 0.700, and 0.734, respectively. The universality and stability of the HTP image data set were verified under objective indices.

Objective: House-tree-person test (HTP) is a projected mental health test based on images. The image data set of the HTP is the basis of intelligent detection and analysis of HTP images. Although HTP is widely used, there are few general data sets.

Theoretical framework: To this end, this study collected images in three ways: Internet download, book scanning, and reality drawing, obtaining thousands of digital images of the HTP. Then, the house, tree and person elements in each image were segmented to form four image data sets of the integral, house, tree and person.

 Method: Therefore, according to the rules, the elements and features of the house, tree and person are individually calibrated, and the experimental data are divided. Finally, the YOLO target- detection model was used for the experiment, and the detection precision of the four data sets under different quantities was analyzed and compared.

Results and Conclusion: The experimental results showed that the average detection precision were equal to 0.828, 0.897, 0.700, and 0.734, respectively. The universality and stability of the HTP image data set were verified under objective indices.