Free AI-103 Computer Vision Solutions Practice Questions

The Implement computer vision solutions domain makes up 10–15% of the Microsoft AI-103 exam. Practice free questions covering image analysis, object detection, custom vision models, OCR, and face detection with Azure AI Vision — each with the correct answer and a detailed explanation.

AI-103 Computer Vision Solutions Sample Questions with Answers

Sample Question 1 — Implement computer vision solutions

A retail company is building an Azure-based shopping app for screen-reader users. When a product photo is uploaded, the app must generate short alt text that helps a person understand what is shown. Which approach is the best fit?

  1. A. Use image understanding or a multimodal model to generate a human-readable description of the photo (Correct answer)
  2. B. Use OCR only and set the extracted text as the alt text for every image
  3. C. Return keyword tags such as colors and objects, and use the tags directly as alt text
  4. D. Use an image generation model to create a cleaner replacement image for accessibility

Correct answer: A

Explanation: Correct answer (A): Accessibility-focused alt text should be a useful human-readable description of what is actually in the image. Image understanding or a multimodal model is the best fit because it can describe visual content in natural language. OCR is only helpful when the key requirement is reading text that appears in the image, and keyword tags are not a good substitute for readable alt text. Image generation solves a different problem and does not help interpret the uploaded image. Why the other options are wrong: - Option B: OCR extracts visible text from an image, but many product photos contain little or no useful text. Even when text exists, OCR alone does not describe the non-text visual content needed for alt text. - Option C: Keyword tags are metadata, not good alt text. Accessibility support should provide a readable description rather than a list of labels. - Option D: Image generation creates or modifies images. It is not the right choice for understanding an existing image and producing alt text.

Sample Question 2 — Implement computer vision solutions

An accounts payable team receives supplier invoices as scans and phone photos. The solution must extract vendor name, invoice total, due date, and line-item tables into structured data for downstream processing. Which Azure approach is best?

  1. A. Use OCR only to read all visible text, then rely on manual parsing for the fields and tables
  2. B. Use Azure AI Document Intelligence for layout-aware field and table extraction (Correct answer)
  3. C. Use image captioning to summarize what the invoice looks like and infer the values
  4. D. Use a multimodal chat model mainly for open-ended questions about the invoice image

Correct answer: B

Explanation: Correct answer (B): This is a structured business-document extraction scenario, not a generic image-description task. Azure AI Document Intelligence is the best choice because it is designed for layout-aware extraction of fields, key-value pairs, and tables from documents such as invoices. OCR alone can read text, but it does not provide the same document-aware extraction behavior. Captioning and open-ended multimodal chat are less suitable for repeatable, production-grade invoice processing. Why the other options are wrong: - Option A: OCR can read text, but the requirement includes structured extraction of fields and tables. Document Intelligence is more appropriate for layout-aware business document processing. - Option C: Captioning gives a high-level description, not precise extraction of invoice fields and tables. - Option D: A multimodal chat model can discuss the document, but the requirement is repeatable structured extraction for downstream processing, which is better handled by Document Intelligence.

Sample Question 3 — Implement computer vision solutions

An insurance portal lets customers upload vehicle photos and ask questions such as, "Does the left headlight appear broken?" The company wants answers based on the actual image rather than on a manually written description. Which design is best?

  1. A. Send the question to a text-only model after an agent writes a short summary of the photo
  2. B. Use a multimodal model in Microsoft Foundry that processes the image and the user's question together (Correct answer)
  3. C. Run OCR on the photo and answer from the detected text only
  4. D. Use an image generation model to create a repaired version of the damaged area

Correct answer: B

Explanation: Correct answer (B): The requirement is natural-language question answering about the actual image. A multimodal model in Microsoft Foundry is the best choice because it can evaluate the visual content and the user's text prompt together. A text-only model depends on an intermediate human summary, which can omit important details. OCR is useful for visible text, not object damage assessment, and image generation does not answer the question. Why the other options are wrong: - Option A: A text-only model can answer only from the written summary it receives. That introduces loss of detail and does not meet the goal of reasoning directly over the image. - Option C: OCR is appropriate when the important content is text inside the image. It is not the best approach for determining whether a headlight appears broken. - Option D: Image generation creates or edits visuals, but the requirement is to interpret the uploaded photo and answer a question about it.

Sample Question 4 — Implement computer vision solutions

A production field-inspection app hosted in Azure calls a Foundry-hosted multimodal model and an Azure storage account. The current implementation stores service keys in application settings. Both target services support identity-based access. What should you do?

  1. A. Commit the keys to source control so all environments use the same credentials
  2. B. Keep using keys, but rotate them during each deployment
  3. C. Use a managed identity for the app and assign the required RBAC permissions (Correct answer)
  4. D. Pass the keys in the prompt so the model can retrieve files directly

Correct answer: C

Explanation: Correct answer (C): When Azure services support identity-based access, the preferred production design is to use managed identity with RBAC instead of storing service keys in code or configuration. This reduces secret-management risk, supports least privilege, and follows Azure-native security practices. Rotating keys is better than leaving them static, but it is still not the best option when managed identity is available. Why the other options are wrong: - Option A: Storing secrets in source control increases exposure risk and is not an acceptable production security practice. - Option B: Key rotation reduces some risk, but it is still weaker than managed identity and RBAC when the services support identity-based access. - Option D: Prompts are not a secure place for credentials, and the model should not receive embedded secrets to access resources.

Sample Question 5 — Implement computer vision solutions

Warehouse workers photograph package labels and serial numbers with mobile devices. OCR accuracy drops mainly on images that have glare, motion blur, or skewed angles. The team suggests switching to a larger multimodal model without changing the capture process. What is the best next step?

  1. A. Improve image capture quality and recapture or preprocess low-quality images before OCR (Correct answer)
  2. B. Replace the OCR step with image generation so the service can redraw each label clearly
  3. C. Keep the same image pipeline and rely on a larger model to overcome poor source images
  4. D. Convert each image into keyword tags first, then infer the missing text from the tags

Correct answer: A

Explanation: Correct answer (A): The problem described is poor source-image quality. Glare, blur, and skew can significantly reduce OCR accuracy, so the best remediation is to improve capture conditions and handle low-quality images before or during OCR. Simply switching to a larger model is not a reliable fix when the visible text is degraded in the source image. Why the other options are wrong: - Option B: Image generation is not the right tool for accurately reading operational label photos. - Option C: A larger model does not reliably solve poor input quality. The root issue is the source image, not just model size. - Option D: Keyword tags do not preserve the exact text needed for serial numbers and labels, so this would not meet the OCR requirement.

Sample Question 6 — Implement computer vision solutions

A manufacturing team wants to check whether a warning light is on in a camera feed every five minutes. They do not need a narrative summary of the full video, and they want to keep cost low. Which design is best?

  1. A. Submit every full video segment to a broad video reasoning workflow
  2. B. Sample selected frames or key frames and analyze those images for the visible condition (Correct answer)
  3. C. Run OCR across every frame to determine whether the warning light is illuminated
  4. D. Fine-tune a text-only model on maintenance notes and use that model for the camera feed

Correct answer: B

Explanation: Correct answer (B): The requirement is periodic detection of a simple visible condition, not end-to-end understanding of the entire video. Sampling frames or key frames and analyzing them as images is the most cost-effective design. Full video reasoning adds unnecessary cost and complexity, OCR is for text, and a text-only model cannot directly interpret camera frames. Why the other options are wrong: - Option A: A full video reasoning workflow could work, but it is not the best fit when only periodic visible-state checks are required. - Option C: OCR is for text extraction, not for determining whether a warning light is visibly on. - Option D: A text-only model trained on notes does not process images or video frames, so it cannot solve this visual detection task.

Sample Question 7 — Implement computer vision solutions

A Microsoft Foundry agent accepts user photos of equipment and can call enterprise tools to open repair tickets or pause a production line. The security team is worried that malicious instructions could be embedded in an image or prompt. Which control set is best?

  1. A. Allow the agent to call any connected tool whenever the model confidence is high
  2. B. Rely only on content filters, because filtered inputs prevent all risky tool calls
  3. C. Restrict tool schemas and permissions, and require human approval for high-risk actions (Correct answer)
  4. D. Disable trace logging so malicious prompts cannot be reviewed after execution

Correct answer: C

Explanation: Correct answer (C): When a multimodal agent can take real-world actions through tools, the secure design is to limit what the agent is allowed to call, apply least-privilege permissions, and add approval gates for high-risk operations. This reduces the impact of prompt injection, embedded malicious instructions, or incorrect model output. Content filters can help, but they do not replace tool restrictions and human approval for risky actions. Why the other options are wrong: - Option A: High confidence is not an authorization mechanism. A model can still be wrong or manipulated while appearing confident. - Option B: Content filters help, but they do not prevent all prompt injection or replace tool-level permission boundaries and approval gates. - Option D: Trace logging supports investigation and governance. Disabling it weakens observability and does not reduce the actual risk.

Sample Question 8 — Implement computer vision solutions

A store operations app must analyze a shelf photo and return JSON with these fields: sku, quantity_estimate, and damaged_visible. A downstream workflow parses the response directly. Which approach is best?

  1. A. Generate a natural-language caption for the image and parse the caption with string splitting
  2. B. Use OCR alone because JSON is just a text format
  3. C. Use a multimodal model with the image and an explicit structured-output requirement (Correct answer)
  4. D. Use an image generation model to create a normalized shelf diagram before analysis

Correct answer: C

Explanation: Correct answer (C): The app must produce machine-consumable JSON from visual content, including non-text judgments such as quantity estimation and visible damage. A multimodal model with an explicit structured-output requirement is the best fit because it can reason over the image and return the required fields in a predictable format. Caption parsing is brittle, OCR alone misses non-text visual information, and image generation is not the right first step for extracting shelf-state data. Why the other options are wrong: - Option A: Captions are designed for human-readable descriptions. Parsing them into JSON is fragile and error-prone when structured output is required. - Option B: OCR can extract visible text such as SKU labels, but it does not by itself determine quantity estimates or whether damage is visible. - Option D: Image generation is not the primary solution for returning structured facts from an existing shelf image.

Keep Practicing

Take the 10-question AI-103 quick-start test across all 5 domains, or return to the AI-103 practice test hub for 200+ questions, exam details, and a study plan.