Is Traditional Image Processing a lost art? : Relevancy check in deep learning era

A surge of deep learning techniques has automated the process of image classification. Traditional techniques of image classification are essentially categorized in two different steps, namely, pre-processing of image data and post processing techniques. Pre-processing involves processing of the image data for robust descriptor definition which are to be used as input for the post processing techniques involving assorted classifiers. However, the aforementioned stages apparently seem to be irrelevant due to automated feature representation of deep networks followed by classification. This talk has explored the significance of handcrafted techniques in the era of automation with deep learning and has attempted a fact check to verify the panacea status of neural networks. Hence, the talk distinctively includes:

• Components of traditional image processing

• Hedgehog Concept

• Crawl-Walk-Run Approach

• Deep Learning Panacea

• Efficient descriptor definition

• Bottlenecks of different approaches