piXserve™ is a client/server application that automatically creates a searchable index of the visual contents of media files. piXserve “sees” the contents of digital images and videos, and stores a description of this content in an XML database. piXserve can also recognize objects, text, and faces of interest to users and can automatically add searchable tags to images and videos where those items are seen by the software.
Through a web browser interface users can search and retrieve images and video segments that contain visual objects that are the same or similar to those in their query pictures. Users can also select specific items in their query image in order to search just for those. piXserve is also able to separate out what is “background” and what is “foreground” in a still image or individual video frame, and these parameters can be used to help formulate more precise search queries. Further, users can formulate search queries using more than one image as an input. In piXserve a search query is encapsulated in a search “module”, and multiple “search modules” can be combined to describe the full query, each module related to the others by means of standard logical operators such as “AND”, “OR”, and “NOT”. Combined, these features provide an unparalleled level of flexibility and specificity to users (ex: “find me an image/video segment where the car in this picture AND the face in this other picture are both visible in the same shot”).
piXserve can also read text that may appear in the field of view of the image, and do so in a variety of languages and character sets (latin characters, chinese ideograms, cyrillic characters, and many others). Users can type a text string and retrieve images/video sequences where such text appears in the image (for example, text on the screen of a news broadcast, or text on a street sign). piXserve can see and recognize text in complex color images and videos at a level of accuracy that is unmatched in the industry.
The piXserve visual search capabilities are complemented by an unparalleled level of auto-tagging capabilities. When the software indexes images/video frames, it tries to add as many keywords/tags as it can to the records that it creates describing the content of the images. piXserve creates keyword descriptors using four mechanisms:
- Face Naming
Users can give a name to a face returned as a search result. When piXserve indexes new images/videos, it checks for the presence of faces. If one is detected piXserve will check to see if it is one of those previously named by users, and if it is it assigns that name as a searchable keyword for that image/video-frame.
- Object Recognition
Users can provide an image as an example of an item of interest, and can give that example a name (a logo, a specific can of soda, a bag of potato chips, a specific car make/model, etc.). During indexing piXserve will check for the presence of an item of interest, and if it is detected it will add a keyword record for that image/video frame.
piXserve can recognize certain generic concepts that may be visible in the image (we call these “notions”). A library of “notions” is included in piXserve, and more are being created all the time. Notions cover a broad spectrum of concepts (for example: flower, car, vegetation, sky, beach, road, map, building, face, bikini, etc.). Users can search for images that contain specific notions or combinations of notions.
- Keyword Recommendations
piXserve can suggest keywords that could pertain to a picture based on the degree of similarity between the contents of that image and those previously indexed by piXserve and for which keyword information is already available.
Combined, the set of features available in piXserve will help your enterprise achieve significant reductions in the cost and time you spend searching and cataloging photos and videos.
piXserve is available in a range of editions to match customer needs, from small workgroups to enterprise-wide and cloud-based environments.
- Automatic Indexing
Point piXserve to a repository of images and video and automatically index their contents. No manual intervention or data entry required. piXserve “sees” what is in the image and automatically creates records that describe the shape, the position, the size, the color, etc. of the discernible objects in the image. Whatever is in the image is automatically indexed.
- Powerful Search
Through a web browser interface, users login to piXserve, connect to available databases and formulate search queries to retrieve images and specific segments of video files:
- Use an arbitrary image from user’s desktop to search for images/video segments that contain the same or similar items
- Use the mouse to point to an area of the query image to indicate which specific item(s) should be searched for.
- Browse the contents of existing databases, and use any of those images to formulate a visual search query.
- Search images by automatically assigned keywords and “notions”
- Type a text string to search pictures/videos where that string appears in the field of view
- Perform not only simple but also complex searches (Example: type a text string to retrieve broadcast news segments where that word is visible on the screen, and then use a picture to search those results and retrieve segments where both the text string and the desired visual object(s) are in the same frame. Use AND, OR, and NOT operators to combine up to 6 criteria in a single query.
- Search by file name (any text substring that appears in the full path of the image/video file).
- Search by geo-spatial coordinates or other external metadata, if available.
- Suggest keywords to describe the contents of a picture
Prune and save search results for future reference. Search results can be annotated and exported. Reports can be generated detailing, for example, the location and count of objects of interest.
piXserve is a multi-threaded, J2EE scalable application that is suitable for the most demanding implementations. piXserve takes advantage of multi-CPU/multi-core environments. piXserve also takes advantage of CPU/GPU accellerator cards that may have been installed on the server.
A REST-based API package is available to support integrations with third party applications and workflow environments.