Business Challenge
Our client receives a large number of crowd-sourced images. They receive many pictures which are blurry and some of the images which are clear do not have the right subject matter. Manually checking and sorting each picture was an extremely tedious and time-consuming process. So, our clients wanted an efficient, automated method for segregating the images that were blurry or had incorrect subject matter.
Approach and Solution
We decided to tackle this problem in two parts:
- Image quality identification
- Image subject identification
Image quality identification
- By creating a machine learning model, we were able to identify whether an image is blurry or not almost instantly. The model takes in various attributes of an image and based on subjective evaluations already done, we trained a model to classify images as blurred, unsure, or good. The unsure images can again be subjected to human evaluation for confirmation. This approach largely reduces the number of images to be manually checked.
Image subject identification
- Next, we created a web application that identified and labeled images according to their subject matter using google vision API.
- Through machine learning and flask, we were able to create an application that can identify the quality of an image and also identify and label the subject matter present in the image.
Benefits
- Two in one solution (Image quality and subject matter).
- Automated quality control is achieved through machine learning.
- Saved 90% of Manual inspection time through automation.
- Segregation of good quality images helped in performing next-level tasks like OCR and Object detection with better accuracy.
Network
Capacity Planing
Our client had thousands of servers spread across the globe carrying internet traffic. The main challenge was managing these servers at optimal utilization.
Vendor Contract Analytics
Our client had thousands of contract documents scanned and stored as pdf documents. Manually scouring through these documents...
CDN
Log Analytics
Our client receives a large number of crowd sourced images. They receive many pictures which are blurry and some of the images...