Spatial Structure in Images
Understand why the positions and relationships of visual features matter.
Spatial Structure in Images explained visually
Compare local details with the global arrangement of a complete scene.
What you will learn
Follow filters, pooling, flattening, and classification through complete CNN examples.
MEANINGFUL ARRANGEMENT
Spatial structure describes where visual elements appear and how neighboring elements relate.
LOCAL STRUCTURE
Local structure includes edges, corners, curves, and textures inside a small region.
GLOBAL STRUCTURE
Global structure describes the broader arrangement of objects across the complete image.
SPATIAL RELATIONSHIPS
CNNs preserve two-dimensional layout while connecting local evidence with scene context.
Quick check: test your understanding
What is local spatial structure?
Patterns and relationships inside a small image region.
What is global structure?
The broader arrangement of features across the scene.
Why preserve layout?
Position and spatial relationships carry useful meaning.
Accessible referenceLesson 42 transcript319 words · Matches the published narration and captionsOpen transcript
Imagine looking at a human face. You recognize the face partly because its features follow a meaningful spatial arrangement. The eyes normally appear above the nose, and the mouth appears below it. If we rearranged those features randomly, the same eyes, nose, and mouth would no longer form a normal face.
This arrangement is called spatial structure. Spatial structure describes where visual elements appear and how they relate to neighboring elements. Images contain both local and global spatial structure. Local structure describes patterns inside a small region. The boundary between a bright area and a dark area may form an edge.
A group of nearby edges may form a corner or curve. Global structure describes the broader arrangement of features across an image. In a street scene, vehicles appear on the road, traffic lights appear above intersections, and buildings form the surrounding environment. Both forms of structure matter.
Consider an autonomous vehicle. The system must recognize individual objects, but it must also understand where those objects appear relative to the vehicle. A pedestrian directly ahead has a different meaning from a pedestrian standing safely on a sidewalk. The visual category may be the same, but the spatial relationship is different.
Traditional fully connected networks often flatten an image into one long list of pixel values. Flattening early can make the original spatial arrangement harder to preserve. A CNN maintains that arrangement while processing local regions. Its filters examine neighboring pixels and preserve information about where detected features appear.
As information moves through the network, the CNN gradually connects local evidence with the larger scene. This ability makes CNNs useful for facial recognition, medical-image analysis, autonomous driving, and object detection. The important idea is that an image contains more than individual pixel values. The position and relationship of those pixels carry essential information. In the next lesson, we will see how CNN layers transform simple local patterns into complete objects.