ImageSimilarity

Find similar images in dataset for a new image using Resnet50+KNN. In order to solve curse of dimensionality using PCA to reduce the dimensionality of features.

Resnet50(trained on imageNet)

1. Create Resnet50 from keras without toplayer to get convolutional features(2048 dimension) as output instead of the image classification probability.
2. Extract the convolutional features for every images of dataset( Feature_size:[number_images, 2048] ).

KNN(Not a classification issue)

1. Fit the nearest neighbor algorithm to the extracted features from dataset
2. Extract convolutional feature of test image(new), and calculate the distance(image_Similarity) between the test image and each image of dataset.

PCA

1. Apply PCA to extracted features and reduce the dimensions.
2. Fit the nearest neighbor algorithm to the new features