Our image mark matching technology comprises a three part process
- Segmenting – analyses key shapes within an image.
- Determining – identifying key component shapes.
- Matching – determining the likelihood that the image elements will create an image match.
Segmentation
A key challenge for image mark matching is that logos and designs are often embedded inside other graphics or superimposed on other elements. They may also be highlighted, unevenly illuminated, poorly rendered or may have blurry edges.
Segmentation of the ‘baseline’ image mark splits an image into component parts, creating regions of relevance within the image and distinguishing it from its background.
Determination
Following segmentation, various advanced heuristics are used to determine if a segment is “relevant”. If a segment is only slightly different from its parent, then the segment is disregarded. This happens commonly with anti-aliased image marks, where the edges of the logo are blurred.
Small segments are typically rejected since they are likely to be seen as noise in the image. Relevant segments are mapped to a grid and stored in a database.
Segmentation of the ‘baseline’ image mark splits an image into component parts, creating regions of relevance within the image and distinguishing it from its background.
Segment Matching
The challenge is to obtain a visual measure of how similar two segments are. Algorithms must be able to reflect “visual similarity”, while, at the same time, must be capable of coping with some variation.
Segment matching algorithms provide a figure representing the “distance” between two segments.
Two segments are said to “match” if their distance is less than a predefined threshold. Each segment is given a weight signifying how important the segment is in the matching process.