Replace backend_converter.py

Now the found images are sorted using the text similarity of the corresponding text segment and the context information of images.