Method for identifying target based on dimension reduction local feature descriptor

One of the most important steps in developing a new drug is target identification and validation. Target-based drug discovery has become the prevailing paradigm used by pharmaceutical and biotechnology companies. This approach is appealing because it holds the promise of identifying more efficacious compounds with fewer undesirable side effects. Good target identification and validation enables increased confidence in the relationship between target and disease and allows the exploration of whether target modulation will lead to mechanism-based side effects. Email:marketing@medicilon.com.cn web:www.medicilon.com

The invention provides a method for identifying a target based on a dimension reduction local feature descriptor and hidden conditional random field. The method is to establish a target identification model for identifying an object, and the model establishing process is a process that the model performs supervised training by using a training image as a sample, wherein each object in the training image corresponds to different label values. The method comprises the following steps of: calculating a descriptor vector of SIFT (Scale invariant feature transform) for the training images of different objects, wherein the descriptor vector corresponding to each image forms a corresponding high-dimensional vector set; performing dimension reduction on the SIFT set by adopting a Neighbor Preserving Embedding (NPE) method; and allowing the vector group subjected to dissension reduction and a label of the object corresponding to a source image to form a dualistic group, namely, each image has a corresponding dualistic group, and the set consisting of the dualistic groups can be used as a sample for training a hidden conditional random field model. An identifying process by the model, namely for a set test image comprises the following steps of: calculating the SIFT feature descriptor set of the test image; reducing dimension of the acquired SIFT set by the NPE method; inputting the vector set subjected to dimension reduction to the hidden conditional random field acquired by training; and outputting the final object label serving as an identification result.

The present invention pertains to a method for dimension reduction target recognition local feature descriptors and implicit condition based on random fields. Specifically, it is the local feature field of computer vision and image extraction, dimensionality reduction methods, and hidden conditional random target image modeling and identification method.

One of the most important direction as the field of computer vision target recognition, is the successor to various higher-level processing such as object classification, retrieval, and so on the basis of understanding behavior. There are a number of ways, including: changes in contour detection based on feature modeling based detection, detection of color statistics EM algorithm, based on the detection area and the detection method based on frame difference and so on. The classic method is simple and easy to understand, but its effect is not satisfactory. Use simple feature information is not sufficient to distinguish objects, the improved algorithm which subsequently emerged, as yet there is mutual offset between the characteristics of certain features, so the more successful target identification methods are so far is specific to a scene below.

The local feature extraction method is characterized by the recent rise in the field of computer vision, has been in object recognition, image registration, image retrieval, 3D reconstruction has been widely used. For local feature geometric transformation, lighting invariant transform for noise, occlusion, and background interference have good robustness, and features a high inter-discrimination.

For object recognition task in terms of local feature extraction completed essential step. Extracting local feature information includes a feature point and the feature point information corresponding to descriptor information. After also need to match descriptors, matching screening process using probabilistic models to complete target recognition, which does not include the process of establishing the library on the object descriptors. Also in utilizing local features matching and then identify the whole process, must also be used to identify the object surface in physical correspondence.

The present invention provides a promoter and hidden conditional random target identification based on dimensionality reduction method described in the local features. It first image extraction SIFT (Scale invariant feature transform, scale invariant feature) feature descriptor, while maintaining the SIFT descriptor of high-dimensional space in relation to the premise, the use of neighbors remain embedded (Neighbor Preserving Embedding, NPE) method for high Victoria descriptor dimensionality reduction, an implicit CRFs (Hidden Conditional Random Fields, HCRF) model and for target identification.