Text detection in images in Computer Vision - Model Pipeline Trace

Model Pipeline - Text detection in images

This pipeline finds where text is located inside pictures. It looks at the image, finds areas with letters or words, and marks them so we know where text is.

Data Flow - 6 Stages

1Input Image

1 image x 640 height x 640 width x 3 color channels→Load and resize image to fixed size→1 image x 640 height x 640 width x 3 color channels

A photo of a street sign resized to 640x640 pixels

↓

2Preprocessing

1 image x 640 x 640 x 3→Normalize pixel values to 0-1 range→1 image x 640 x 640 x 3

Pixel values changed from 0-255 to 0.0-1.0

↓

3Feature Extraction

1 image x 640 x 640 x 3→Apply convolutional layers to detect edges and shapes→1 tensor x 80 x 80 x 256 features

Edges of letters and shapes highlighted in feature maps

↓

4Text Region Proposal

1 tensor x 80 x 80 x 256→Detect possible text areas using bounding box proposals→1 tensor x 80 x 80 x 5 boxes

Boxes around areas that might contain text

↓

5Bounding Box Refinement

1 tensor x 80 x 80 x 5 boxes→Adjust box positions and sizes for better fit→1 tensor x 80 x 80 x 5 refined boxes

Boxes tightly fit around text regions

↓

6Non-Maximum Suppression

1 tensor x 80 x 80 x 5 refined boxes→Remove overlapping boxes to keep best ones→Variable number of boxes (e.g., 10 boxes)

Final boxes marking text areas without overlap

Training Trace - Epoch by Epoch

Loss
1.2 |*       
0.9 | *      
0.7 |  *     
0.5 |   *    
0.4 |    *   
    +---------
     1 2 3 4 5 Epochs

Epoch	Loss ↓	Accuracy ↑	Observation
1	1.2	0.45	Model starts learning to detect text regions
2	0.9	0.60	Loss decreases as model improves detection
3	0.7	0.72	Model better at finding text boxes
4	0.5	0.80	Accuracy rises, loss continues to drop
5	0.4	0.85	Model converges with good detection performance

Prediction Trace - 6 Layers

Layer 1: Input Image

Layer 2: Preprocessing

Layer 3: Feature Extraction

Layer 4: Text Region Proposal

Layer 5: Bounding Box Refinement

Layer 6: Non-Maximum Suppression

Model Quiz - 3 Questions

What is the purpose of the Non-Maximum Suppression step?

A To normalize pixel values

B To remove overlapping boxes and keep the best ones

C To resize the input image

D To extract features from the image

Key Insight

Text detection models learn to find areas in images that contain letters by first extracting important shapes and edges, then proposing and refining boxes around these areas. Training improves the model by reducing errors and increasing accuracy in locating text.