Best Visual Search Engines Review: (2) is one of the earliest visual search engines. It starts from another visual search engine called Riya, which was originally used to search celebrity people, objects in the pictures, and faces in the photos. From the profitability point of view, now changes its business model to high-end online shopping based on visual search. 

The first thing I would like to talk about is’s business model. There are always cool computer vision technologies. However, how to leverage the technology to make profit is a different thing. is undoubtedly a good model for visual technology entrepreneurs. Visual search is high-tech, and online shopping is something popular and something you can make profit from. The marriage of high-tech search engine with online shopping will undoubtedly bring you venture capital and eventually gain profit for you. 

In terms of visual search technology and functionalities, predefines many search categories (e.g., clothing / outwear / swimshop / sleepwear / maternity, and there are subcategories under each category). Your search can be refined by style (shape), color, brand and/or site. It also has “product search by image” functionality. You upload a photo of your favorite products (e.g., handbags used by some celebrity), you would like to find products with similar color, pattern (texture) or style (shape) on the web. As we know, this task is very challenging in technology and time-consuming to index/search all the images. provides a limited number of predefined product categories and styles. A user needs to select his/her product category and style before uploading a photo. You can follow the six steps to find what you want.

Step 1: Given the following image, you select category (Women’s Handbags) and style (Tote). 

Step 2: Upload your photo from your local computer

Step 2: Draw a box around your item. A simple user interface is provided for you to crop your object of interest out. 

Step 4: Refine your search by Color, Shape or Pattern of the object. “All three are equally important” is selected here. 

Step 5: Enter Keywords, such as brand, materials (leather, woven, etc.). For example, you can input “leather” in “Keywords”. 

Step 6: Enter your email. Obviously this is still a beta version of its search-by-image functionality. 

And then click the “Start the search” button, you will get the following message:

You’re done! We’re analyzing your photo and sorting the results by color, shape and pattern. Keep an eye out for an email from with your search results. Also, make sure to add to your address book. In the meantime … Browse for Tote in Women’s Handbags.

This is a little bit surprising. One reason a user wants to leverage this kind of search is because of its instant feedback. However, it took about one to two hours before I received the email from informing the search results. 

From the results, we can see that it did find similar handbags from different commercial websites, though some of them are arguable. Interestingly, there are two lines of search tips at the end of the search results page, reading as:

The results are only as good as your original photo. For the best visual matches, use the following tips: The background should be a solid color with no arms or hands holding the bag

Did it recognize that the input image does not have a solid color, and there are arms or hands holding the bag? This is a much harder problem than recognizing the handbag itself, from the technology point of view. I would rather believe that any cluttered scene for searching a Tote handbag would generate such a tip. 

There are at least two possible reasons why did not provide instant results. 

  1. The speed of the visual search engine is unbearably slow. This is less likely because the number of images for each category and style should not be large. Even off-the-shelf algorithm should work pretty well on this small dataset. 
  2. Some manual work is involved. Considering not many people are using their search engine at this stage, it is possible that the search results were manipulated by somebody at the back end. 

Following the search tips, we then used PhotoShop to manually segment the handbag out. We did further tests by rotating the image to the following three orientations. 

Similar results were returned for these three cases. Therefore,’s search engine is rotation invariant because shape information is used. 

We are very glad that has given some try to the challenging problem of image search by shape, pattern and color. The search results are reasonably good. Mostly importantly, it finds its way to the market. 

The overall evaluation of is as follows:

  • Search functionality: 
  • Search accuracy: 
  • Search speed: 
  • Database completeness: 
  • Technologies:
  • Business model: 

No Comments

Be the first to comment!

RSS feed for comments on this post. TrackBack URI

Leave a comment