Image Annotation of Palm Images for an Astrology App

The Client

A Tech Company Developing an Astrology App for Palm Reading

Our client is a technology company creating an AI-based application for palm reading. The app allows users to upload photos of their palms to the app, which then uses artificial intelligence to analyze the lines and features. The app compares these results with established astrological guidelines to generate personalized readings, offering users a seamless and personalized experience.

PROJECT REQUIREMENTS

Image Annotation for Accurately Labeling Palm Images

The client approached us with a critical task to provide accurately labeled palm images for training their AI algorithm. The specific requirements were:

A large dataset containing a total of 10,000 palm images were to be labeled, for AI model training.
Precisely identify and categorize various regions of the hand, such as finger points (5 regions), the area below the fingers (5 regions), and hand mounts (9 regions per palm).
Ensure high-quality, consistent labeling across a diverse set of hand images to maximize the AI's ability to identify these critical regions when processing user images.
Deliver labeled images in a format compatible with the client's AI training pipeline.

PROJECT CHALLENGES

Addressing Posture Variations and Unrecognized Areas for Accurate Palm Lines Recognition

The project involved significant challenges that required our team to develop custom solutions and refine our processes., including:

Diverse Images: The palm images received by the client varied significantly in terms of postures and positions, making it difficult to identify key points like hand mounts. Additionally, variations in lighting, hand angles, and finger positioning further complicate the annotation process.
Unrecognized Palm Areas: While the core focus was on labeling key regions like finger points and hand mounts, there were additional unrecognized areas, such as gaps between fingers which had to be carefully categorized by annotators.
Image Clarity Issues: Several client-provided images were unclear which affected the ability to recognize the palm lines. This was mainly due to poor lighting, darker skin tones, or shadows.
AI Tool Limitation: The tool LabelBox was insufficient for accurately detecting and labeling certain regions, such as the gaps between fingers or the edges of the palm. Manual intervention was necessary to accurately label these regions.

SOLUTIONS OFFERED

Leveraging Annotation Tool and Human Expertise for Accurate Palm Image Labeling

We allotted a team of 10 dedicated image annotators for the project. To overcome the challenges and meet project requirements we implemented the following solutions:

Use of Labelbox Annotation Tool

Our image annotation team leveraged the LabelBox image annotation tool to label the palm lines and hand regions accurately using polygon and polyline techniques.

Referencing External Sources for Quality Labeling

For accurately labeling palm images, we used verified images from sources like Google Images as a reference guide. This allowed us to standardize the labeling process and ensure that every image was consistently annotated according to astrological guidelines.

Manual Image Labeling

We used manual annotation techniques to accurately identify and categorize areas that AI struggled with, such as the gaps between fingers. Our team used Google images as references for palm lines, ensuring precise and consistent labeling across all images.

Quality Control

We implemented a rigorous in-house quality check process to verify that all required points across each palm image were correctly annotated. This ensured the integrity of data before it was fed into the AI training pipeline.

Image Processing for Clarity

For dark and unclear images, our photo editing experts processed the images by adjusting their opacity and brightness to enhance image visibility and perform the annotation process.

Hand Segmentation

Before
After

Palm Lines

Before
After

Hand Mount

Before
After

Finger Point

Before
After

Bottom of Fingers and Hand

Before
After

Project Outcomes

Enhanced Dataset Quality

Our rigorous QC process and image processing measures ensured that the final dataset was free of errors or inconsistencies. This was a crucial factor in enhancing the AI's ability to perform reliable palm readings and generate responses for users.

Improved AI Accuracy

A precisely labeled dataset enabled a 25% increase in the client's application's accuracy in identifying palm lines, hand mounts, and other critical regions.

Feed your AI Applications with Accurately Labeled Data

Turn to our image annotation services that infuse the human-in-the-loop approach to ensure 100% accurate and reliable labeling for your AI/ML models and applications. To know more about our services or discuss project requirements, write to us at info@suntec.ai

Take AI to Production

Get high-quality training data. Request a FREE sample.

Enhancing an astrology app's palm reading accuracy by 25% with precise image annotation services