Describable Visual Attributes for Face Verification and Image Search

Describable Visual Attributes for Face Verification and Image Search

Most prevailing and standing methods for the recognition work by removing and extracting low-level structures and scores in images. For example: gradient directions, pixel values, histograms of oriented gradients etc. These methods are used by the classifiers for the identification and recognition of the visual attributes. In this research, the use of favorable visual attributes and image search has been describes. Tangibly describable visual attributes are known as labels that can describe the image and the appetence. In this study the approach that has been used is to set a large vocabulary in order to describe the visual images, which is used in order to train the classifiers. These classifiers then will robotically recognize the extent to which these attributes have been present and matched in the new provided images.

The question here arises is that why one would need any of these attributes? Visual attributes are just like the words; they are tremendously flexible and compos able and can be combined in different ways.

This research emphasis is on the provided images of the faces and the attributes that can be used to describe those images, examples of the labels and the face attributes, that are used to define the images include, jaw, gender, shape, size, nose, etc.

In this research we have created and presented & exhibited the large sets of the images from the real world to train the classifiers. During this training we have enabled the trainers to identify the absence, presence and the extent in which any of the attribute is present. And after identifying all the factors these classifiers are able to detect and identify the new images. This training also has explore the future potential and have measure the extent and accuracy in which these face verification attributes can be used to determine the image search through the computational and human researches and experiments.

In the end, the two new data identification sets have been introduces. These are Pub fig and Face Tracer. They provided the labeled and mentioned images with the detected attributes.

Face Tracer Data Set:

It is a sub set; the face tracer data set has been established by the face Data base of the Columbia. And it includes the labeling of the attributes.in this data set, it has 15,000 different faces and this saved data has marked and labeled fiducial points and metadata.

Pub Fig Data Set:

This type of data set has been a complement to the data set known as LFW. It includes around 58,797 images and pictures of the 200 public figures. It is different from the face tracer because it has included a variety of the images of the same person when it has been compare to the LFW. It has constructed the subsets of the data that has been categorized in different lightening or poses or expressions and conditions.

In this research work, it has been shown that how to train and prepare the classifiers to describe and label the visual and image based approaches. They have been trained by using and sowing a number of labeled and identified mages from the internet. This research has also established the use of these describable and tangible attributes for execution and identification of face verification and image exploration.

The advantages of this approach based on the attributes representation are:

1. These attributes can be used to describe and specify the descriptions of the faces at different levels and with specificity.
2. They are common attributes as when learned can be then used to analyze and detect new objects and images without any prior training in different categories.
3. This approach is more efficient and requires lesser attributes to identify

In this research, the performance of this research has been shown in comparison with the other image search engines and applications. Such as, the face verification, face search and attribute classification. This research has also established and made two complementary data sets of the images and have made them available to the community so that  there future progress can be judged and made along these lines. The research and experiments have reflected that by adding and labeling more face attributes and improves and advances this process of face verification in the future and great benefits can be yielded.