4.1.1. Image Classification

Image classification assigns a single label to an entire image. It answers: "What is this image of?" Think of it like sorting photos into folders—each photo gets one folder (label), and you don't care about specific locations within the image.

Key characteristics:

• Output: Single category label for the whole image
• Does NOT provide location information
• Does NOT identify multiple distinct objects
• Used when you need to categorize images into predefined groups
• Confidence score indicates how certain the model is

How classification works:

1. Model analyzes the entire image as a whole
2. Compares visual patterns to learned categories
3. Returns the most likely category with a confidence percentage
4. Example output: {"label": "beach", "confidence": 0.94}

Common scenarios:

• Sorting photos by content type (vacation, work, family)
• Quality control in manufacturing (pass/fail classification)
• Medical imaging triage (normal/abnormal)
• Content moderation (appropriate/inappropriate)
• Product categorization in e-commerce

Classification types:

What classification does NOT do:

• Tell you WHERE objects are located (use object detection)
• Count how many objects exist (use object detection)
• Extract text from images (use OCR)
• Describe the image in sentences (use image captioning)

⚠️ Exam Tip: Classification questions often use phrases like "categorize," "sort," "identify what type," or "determine if the image shows." If the question asks about LOCATION or POSITION, it's NOT classification.