Attention Makers

CATEGORY: ENTERTAINMENT

Movie segment retrieval

MAKERS: Krishna COUNTRY: India

This project aims at retrieving video segments from a movie given a thumbnail or image. This is very useful in content retrieval and it is done using Deep learning technique.

The Purpose

This project aims at solving image based content retrieval using Deep learning technique. For example if i want to see a short clip in a movie which i like, i can just present a image of that clip and the system will search the clip for you. This reduces the burden of searching the whole movie manuallly for a short clip.

The Technology

In this project we Deep Convolutional neural network (19-layer) to extract the high level representation from the movie frames.We extract high level features from the video frames and store them in a matrix. This is like compressing the entire movie into a matrix. During testing we take the image presented to the system and compute the features from the same CNN and compare this feature with the movie feature matrix just by simple dot product. Now the system can find best matching scene corresponds to the image and we play the video for 20 seconds

Additional Details

We use pretrained VGG-19 network to extract the features. VGG network is trained on ImageNet dataset. We tap into the last fully connected layer to extract the image features which are 4096 dimension real numbers.

Home Previous Next

Vote Share Comment