Cross-modal Image-Text Retrieval Using Support Vector Machine

With the increasing growth of multimodal data in the form of audio, video, image and text data, the importance of multimodal retrieval has also increased. One of the main challenges of cross-retrieval is to reduce the heterogeneity gap between different methods, such as retrieving images through texts or vice versa. Therefore, in this paper, a reciprocal retrieval method based on supervised learning is proposed. Image features including color, texture and shape are extracted using color auto-correlogram, Gabor filter and Zernike moments. Text features are also extracted using latent Dirichlet allocation method. Also, two support vector machines are trained separately to learn the features of images and side texts. Finally, mutual retrieval is done based on the classification results of the search modality and considering the smallest distance between the samples of the opposite modality.

Keywords:

Cross-modal retrieval Support vector machine , Auto correlogram , Gabor filter , Latent Dirichlet allocation