Improving Referring Image Segmentation using Vision-Aware Text Features