Abstract
Since most multimedia database systems are built on top of object or object-relational database systems, they inherit the underlying query facilities. The approach we present in this paper is in two steps. The first step is to design a multimedia query language that will be used as an internal language. The second step is to define an equivalent visual query language and a translator to translate a visual query into a query in the internal query language.
Introduction
Presented at the Visual Database Systems 4 conference in 1998, VisualMOQL is a graphical interface designed to make querying complex image databases as intuitive as drawing a sketch. By acting as a bridge between human-friendly visuals and a powerful internal language called MOQL (Multimedia Object Query Language), the system allows users to perform sophisticated searches without needing to write a single line of code.
In the late 90s, a significant gap emerged between "naive" users and the database systems they needed to search. While object-oriented systems had robust textual query languages like OQL, they were often too complex for average users. Conversely, existing visual query languages lacked the expressive power for complex logic. VisualMOQL offers the ease of a graphical user interface (GUI) with the underlying "muscle" of a formal, object-oriented query language.
System Architecture and Methods
The researchers followed a two-step approach: developing MOQL, an extension of the standard Object Query Language (OQL), and building VisualMOQL, the visual layer that translates user actions into MOQL.
Key components of the system include:
- The DISIMA Model: A framework that separates the "image layer" (the concept of the image) from the "representation layer" (the actual file/data), ensuring representation independence.
- Logical vs. Physical Salient Objects: The system distinguishes between a physical object (pixels) and a logical object (the abstract concept, e.g., "Bill Clinton").
- The Working Canvas: Users select object classes and draw rectangles. The system automatically calculates topological and directional relationships (e.g., "east" or "west") based on centroids.
- The Query Canvas: A higher-level space where users combine simple sketches using Boolean operators (AND, OR, NOT) to build complex compound queries.
Query Translation and Results
The core innovation is that the visual language acts as a translator for a robust textual language, allowing the GUI to inherit the power of the underlying Database Management System (DBMS). By using simple rectangles and their centroids, the system programmatically defines spatial predicates like "next to" or "above."
The authors demonstrated that complex natural language queries—such as "Find images with Bill Clinton dressed in black next to Jean Chrétien OR images with Hillary Clinton sitting next to Jean Chrétien"—could be represented visually. The system's translator turns these visual arrangements into multi-line MOQL statements.
As seen in the translation, the system handles:
SELECTandFROMclauses identifying relevant image and person classes.WHEREclauses defining object names and spatial predicates (e.g.,P202 east P201).- Color similarity functions (e.g.,
color colorsimilar(0.5) (0, 0, 0)).
Limitations and Future Work
While robust, the prototype has limitations. Salient objects and attributes are identified during "database population time," requiring significant manual or semi-automated tagging. Additionally, while MOQL supports temporal expressions for video, the VisualMOQL interface primarily focuses on images. Extending these visual metaphors to the time domain for video search remains a logical next step.
Summary updated