Build an End‑to‑End Image Search Service with Alibaba Cloud OpenSearch Vector Retrieval and AI Search

This article introduces how to combine Alibaba Cloud OpenSearch Vector Retrieval with the AI Search Open Platform to build a one‑stop image search service that automatically handles image vectorization, OCR text extraction, and similarity search.

Background

In the digital era, demand for image search is growing across e‑commerce, media, and logistics, but traditional keyword‑based search cannot handle unstructured image data.

Vector Retrieval

Vector retrieval converts images into high‑dimensional vectors and measures similarity by distance, enabling efficient image matching and text‑based queries.

Typical Issues

Insufficient technical expertise to process images into vectors.

Traditional visual search yields poor results for similar‑looking items.

High maintenance cost as data volume grows.

Solution Overview

The solution uses Alibaba Cloud OpenSearch Vector Retrieval for indexing and searching, while the AI Search Open Platform provides OCR and image‑embedding models. The workflow includes data preprocessing, index creation, and query testing.

Implementation Steps

Data Preprocessing : Upload raw images; the platform generates source_image_vector and source_image_ocr_vector fields.

Index Configuration : Choose the "Vector: Image Search" template, set table name, shard count, and resource limits.

Field Mapping : Use default fields id, source_image, namespace, source_image_vector, and add OCR vector field.

Vector Index Settings : Define vector dimension, distance type, and indexing algorithm.

Data Ingestion : Add image records via developer mode (JSON) or form mode (single record).

Products Used

AI Search Open Platform : Provides OCR and image‑embedding services.

Alibaba Cloud OpenSearch Vector Retrieval : Offers scalable vector indexing and similarity search.

Operation Manual

Pre‑conditions

Registered Alibaba Cloud account with real‑name verification.

Access‑key created for API authentication.

VPC environment ready.

Step 1 – Purchase Instance

Log in to the OpenSearch console, select the Vector Retrieval edition, choose specifications, and complete the purchase.

Step 2 – Enable AI Search Platform

Activate the AI Search Open Platform service, create an API key and domain.

Step 3 – Configure Vector Instance

In the instance list, set basic table info, select the "Vector: Image Search" template, configure shard count, resource slots, and confirm creation.

Step 4 – Field Configuration

Map source_image to the image preprocessing template (choose Base64 or OSS), enable OCR and embedding models, and generate the vector fields.

Step 5 – Index Creation

Create separate vector indexes for image vectors and OCR vectors, set dimensions, distance type, and algorithm.

Step 6 – Data Import

Use developer mode (JSON) or form mode to add image records; required fields are id, source_image, and namespace.

Step 7 – Query Testing

Perform searches in form mode (image upload or text input) or developer mode (API). Distance scores indicate similarity (smaller is better).

Key Benefits

The integrated solution lowers technical barriers, enables fast deployment of image‑search capabilities, and improves query accuracy and response speed for enterprises without dedicated AI teams.

Images