Our current service may not always satisfy all users at the first query. Human eyes are very sophisticated systems and it's a tough job for computer vision to catch up with them. Here are tips when you encounter search results you don't expect.
1. What you can do
1) Search Option
Try search options ; shape, layout and color.
2) Face Search
At current user interface, face search are not active at the first query. If you want to find same or similar faces, set 'face' option.
3) Combined Search
Enter any keywords so that combined search, image plus keyword search, works.
2. What we can do
1) Increase database
We think the more image data we have, the better results we can return. Our search is based on similarity, so results are affected by the size of the database. I experience the results from 10 million images are much better than that from one million images. We're continuously crawling the web and the searchable data is increasing. Try our service after some interval.
2) Limit types of query
Since our database consists of various images from the web and our algorithm purely rely on color and shape, different types of images are displayed as similar results. If type of database and queries are same, more relevant results can be returned. We're preparing new service that has specific database and I think search results would be more close to what you are expecting.
3) Improve image recognition
It's ideal to develop algorithm that can detect everything in images. I don't know how many years, decade and centuries are required to make it come true, but technology is moving forward.