The onset of visual search tools has fundamentally changed the way users seek information online. While text keywords have been, and will continue to remain, the first choice for users in describing their search engine criteria, voice and visual search tools are catching up fast. The latter is expected to comprise nearly 50 percent of all Web and mobile searches within four years, according to Baidu, the world’s number-two search engine. These trends stem from basic human preferences in exploring data. Our brains are wired in such a way that we respond better to visual stimuli and voice than to having to describe complex ideas in words.
Let’s take a business case example to illustrate the advantage of image-based search over text. Recently, I needed to purchase a wireless lapel microphone system for an important trade show but was unable to describe the exact product configuration in words. Soon I checked into a mega audio equipment store, which boasted a rather huge collection of sound systems – speakers, microphones, noise cancelling headphones, you name it.
It did not end my quest though. The sales rep, a very helpful person, was unable to follow the precise requirements I gave him. I showed him a couple of pics to see if he could relate better. The company carried the products, but the sales rep wanted me to describe them correctly using exact keywords as, otherwise, it would lead to an endless search in a company catalog featuring thousands of products.
Well, e-commerce companies recognize struggles like this on a daily basis. It’s much easier for people to click a picture of their desired product, upload it on an image-based search engine and arrive at close matches, rather than using text to describe every single product attribute.
It’s safe to assume that visual search engines (and the twin emergence of camera phone technologies) have significantly bridged the divide between end-user search capabilities and their object of desire. For e-commerce companies bound by the universal KPI of higher site conversion rates, visual search tools can greatly help grow their business. But it is becoming an unrealized territory.
Most e-commerce experts agree that visual search technologies haven’t been realized to their full potential so far, mainly due to still-persisting gaps in technology. For one, most visual tools online haven’t really evolved in their algorithms in terms of differentiating objects accurately from their surroundings. The signal-to-noise ratio is way too high. This is a problem that may be solved using computer vision segmentation techniques (technically known as content-based image retrieval or CBIR), the same methods used by forensic experts to separate valid clues from misleading trails in a scene of crime. Attributes such as color, shape, size and proportions may be more accurately described using CBIR compared to other methods.
By implementing image-based search engines on Web and mobile apps, e-commerce companies should fundamentally achieve the following results:
- Greater autonomy for end users. From a user perspective, the number of search steps needed to arrive at the end product should be minimal.
- Image search engine results should be reusable after every search.
- Fast and accurate performance of the search engine. The response to search queries and sorting of results should take less time. This can be achieved with integration in NoSQL databases. Depending on the complexity and size of product catalogs, one requires expertise in using tools such as MongoDB, OpenCV and Python. Large-sized retailers are better off roping in outside experts for data analysis requirements.
Sayak Boral is assistant manager of marketing at eInfochips, a Sunnyvale, Calif.-based product development and software R&D services company, where he handles corporate communications, branding, PR and content development for next-generation technologies such as visual search engines, advanced QA practice, Internet of Things and more. He has more than eight years of experience in diverse industries including semiconductors, enterprise IT, telecom OSS/BSS, network security and core software. Contact him at firstname.lastname@example.org and connect with him on LinkedIn.