Pathologists rely on histology information for disease diagnosis and patients treatment. However, the process of disease characterization by visual examination of histology tissue images is often labor intensive and requires expert knowledge. In this dissertation, we present algorithms for automatic identification of various biological structures at different scales, for example, tissue regions, immune cells, and subcellular structures. These biological structures of interest are closely related to the diseases, and are needed to be identified first, so that properties of these biological structures can be extracted to analyze the diseases. However, in histology tissue images, these biological structures usually have large variations with respect to, for example, shape, size, and color, both across different images, or within the same image. Also, the biological structures are embedded in a very complex tissue background, which is quite noisy and possibly has considerable artifacts. Therefore, it is very challenging to identify these biological structures. The proposed algorithms are based on combinations of computer vision and machine learning (including deep learning) techniques, to resolve or alleviate the challenges. As a result, the proposed algorithms not only provide useful tools for automatic identification of biological structures (which greatly releases the burden of visual examination of histology tissue images from pathologists), but also introduce an opportunity for quantitative analysis of the diseases (which substantially reduces the degree of subjectiveness of disease analysis, and provides novel insights into the diseases).