Autoencoder for Image Retrieval System using Deep Learning Technique with Tensorflow and Kears

In today's era image retrieval is most emerging trends of digital image processing techniques. Traditional methods for retrieval such as Text and Query based image retrieval of image it will take more time for searching and retrieving images from large datasets. So, to reduce the time for searching and results, in this research work we used deep learning neural network technique is called Autoencoder. Autoencoder is neutral networks which reconstruct images. The proposed work, the image are retrieves similar visual content textures, color and shape of image this technique recognized as “Content Based Image Retrieval (CBIR)”. Current research work, we used the autoencoder technique to form image into feature encoded vector format which will helpful for searching image from larger dataset. Proposed the encoder architecture to evaluate the input image then applied two training approaches pre-train and post train to generate good quality encoded representation. Three autoencoder are used for this research work Simple autoencoder, Deep autoencoder and Convolutional autoencoder. To reduced search indexing time, for extracting similar images from the dataset we used Approximate Nearest Neighbor (ANN) approach. Finally, developed end to end application system for image retrieval on MNIST and CIFER 10 datasets.