The image search engine recognizes the name of the object itself

Technology of the University of California (USA) will automatically put multiple names for images in words corresponding to what it "sees" like " umbrella ", "football player" or " model ". Every part of the image is scanned and conceptually attached.

According to the researchers, this method will ensure accuracy when searching for images because every keyword can be met in a maximum way. Currently, the search algorithm only scans the keywords in the accompanying text, such as the name of the image and the article containing the image.

This technology can identify images as "person", "water" (water) or "swimmer" (swimmer).Photo: Technology Review .

The new technique will describe objects in a picture by analyzing features such as colors, contours, textures . Each object is presented in many ways and will be compared to samples taken from the picture. other photo. The computer will associate the correspondence with this parameter to return the results to the searcher.

However, Google expert Pedro Moreno thinks that this way takes more intermediate steps. For example, if you want to upload a picture of a dog lying on the grass, the software will have to scan through the categories to distinguish the dog from the cat or the tiger and the leopard. After that, it will give an analysis to get the images corresponding to these parameters. Up to 60% of the pictures are dogs, the rest are cats or other animals. Meanwhile, the system also scans thousands of other images with a background of mountains, houses, flowers . to find out where the grass is.

Larry Zitnick, an image search expert at Microsoft, also said that this way is difficult to handle large image systems on the Internet because it takes a long time to analyze. " However, this is also a good way for some simple images in paintings ," he said. " Image problems are complex and there is no single solution that meets all requirements ."

Google said that this method will be useful when integrating with existing search software. It is appropriate when people look for specific images, not abstract things like " sad " or "happy." For example, the need for a picture of a happy family with parents, a son and two girls walking in the evening seems beyond the capabilities of current technology.