Computer Vision Assignment

Applications of Computer Vision on Mobile Phones

• Face recognition

  Used to unlock the phone and to protect apps with biometrics.

• Object and text scanning

  Apps (for example, Google Lens) read text, translate, and identify plants, products, or landmarks from the camera view.

• Image and video enhancement

  Social apps use detection of faces and scenes to add filters, remove background, or improve image quality.

Steps in Object Identification (Photo of Desk)

1. Image input User uploads or takes a photo.
2. Visual scan The model finds edges, shapes, colors, and patterns.
3. Object recognition The model matches detected features to known objects in its training data.
4. Output generation The system lists labels, names, or short descriptions of recognized items.

Objects Likely Identified on a Study Desk

• Open textbook with anatomy diagrams
• Pens, markers, and pen holders
• Sticky notes
• Lamp and small clock
• Glass jar
• Laptop
• Shelves with spiral notebooks
• Wall notes and study diagrams
• String lights

Challenges for ChatGPT and Computer Vision Systems

• Small or unclear text Handwriting and small print are hard to read.
• Poor lighting Shadows or warm light change colors and hide details.
• Object overlap Items stacked or close together can be counted or labeled wrong.
• Ambiguous shapes Containers, jars, or cups may look similar and be confused.
• Context limits The model does not know the purpose of an object without extra information.

Highlighting or Circling a Specific Item

When asked to circle an item (for example, "circle the clock"), the system can draw a circle on the image. Common issues:

• The system may choose the wrong object if several objects share the same shape (for example, two round items).
• Circling may cover or hide small nearby objects.
• The drawn marker may look large or imprecise on small targets.

Untrustworthy or Unsafe Applications of Computer Vision

1. Facial recognition in public surveillance

• Can misidentify people, especially in poor light or for darker skin tones.
• Can track people without consent and harm privacy.
• Has been used to monitor activists or arrest people unfairly in some places.

2. AI-based hiring tools

• Some systems analyze faces, tone, or body language in interviews.
• These tools can be biased and may favor certain races or genders.
• They cannot measure true skills such as teamwork or kindness from appearance alone.

3. Deepfake generation

• Computer vision can create fake videos that look real.
• Deepfakes can spread false information or be used to harm reputations.