Description
For car make identification, we first came up with using the SIFT descriptor and SIFT matching with RANSAC. The matching results of two images from Hyundai Sonata are shown in Figure 1.
From Figure 5, we observe that only a very few points get matched while those points don’t indicate the same feature on the vehicle. SIFT matching is robust with matching among planes. However, it cannot handle 3D image matching.
We implement a simple convolutional neural networks based on Tensorflow. Our networks have 3 convolutional layers. Each convolutional layer comes with a max pooling layer to do down-sampling. The filters we use in each layer are specified in the figure 10. We also use a softmax classifier to output class scores.
For the project, we only implement two simple neural networks. They give relatively good result compared to SIFT matching. Since we only have 100 training images for each car make, the trained networks might be biased so that the accuracy stays around 75%~78%. We believe the accuracy can be improved by using larger datasets. Besides, in our results, a regular neural networks and a convolutional neural networks both achieve satisfying results. However, we still consider convolutional neural networks a better method to do car make identification. Regular neural networks will fail to handle huge number of parameters and lead to overfitting when input images are large . With larger datasets and deeper convolutional neural networks, our proposed method can be robust and accurate in car make recognition.
Python code for merging two separate photos to create a single, multi-person photo
Python code take a a binary image which fed to the Tesseract OCR
Python code for Classification of photographic images based on perceived aesthetic quality
Reviews
There are no reviews yet.